Abstention

An AI assistant that calls tools makes two decisions on every request: which tool to invoke, and whether any available tool applies. In the usual design a single language model makes both, by emitting a call or by declining to emit one. On a device that has to answer without a server, the language model is what makes that design expensive, dominating both the latency and the memory of the router. The common alternative is to remove the model completely and rank the catalog of local actions with a retriever instead. That substitution is not symmetric across the two decisions. A retriever returns its highest-scoring candidate for every input and cannot signal that the catalog holds no valid action. Our earlier study found that constraining a decoder to a tool grammar repairs malformed output without improving the choice. What the substitution costs in each decision has not been measured. We evaluate the two decisions separately over 600 Korean and English requests and a catalog of 70 local actions. The router may also ask for a missing slot, reply, or delegate. Half the in-catalog requests reuse catalog vocabulary and half paraphrase it, separating lexical overlap from the action requested. Character 3-gram BM25 selects 162 of 164 lexically matched requests and 85 of 166 paraphrases. Restricting the candidate set to seven raises the paraphrase figure to a mean of 0.825 over five trials. No classifier over its score features separates in-catalog from out-of-catalog above 0.697 area under the curve, where the frozen encoder multilingual-e5-base reaches 0.806. Using that encoder for abstention alone keeps 376 of the requests local and misroutes 9 of the 150 needing delegation. Abstention, not selection, is where a neural component is required. A neural ranker improves every quality metric and is rejected on latency and memory rather than accuracy.

Three recent results describe what look like unrelated LLM reliability problems. Yin et al. (2026) show reasoning RL collapses tool-reliability representations. Suleymanov et al. (2026) show that under safety-constrained generation, large models rewrite flagged spans while small models truncate. Bastounis et al. (2024) prove any consistent-reasoning system without an implicit "I don't know" function must hallucinate infinitely often on broad problem classes. We argue these findings converge on a single intervention: calibrated abstention is what each independently identifies as the missing capability, even though the unavailability they document, a capability gap, a policy gap, and a recursion-theoretic gap, has a different source in each case. Honesty post-training has narrowed the gap in deployed models, but principled closure of the class Bastounis identifies requires a calibrated abstention function whose training signal at the leaderboard level is absent: dominant benchmarks assign zero reward to decline, so the leaderboard gradient that would select for the function does not exist. We propose four changes to evaluation: triple-scoring, abstention-rate reporting, capability-stratified evaluation, and mandatory calibration metrics. Benchmark reform is necessary, not sufficient, for closing the gap the theorem identifies.

Large language models can produce fluent answers when their factual support is weak. This paper introduces Chain-of-Self-Questioning (CoSQ), a prompt-only framework that makes answer commitment conditional on an explicit assessment of the information required to answer a question. We evaluate three CoSQ variants under seventeen conditions on the 817-item TruthfulQA multiple-choice validation set using eleven open-weight and hosted model families. In the final balanced-option protocol, Grounded-CoSQ at τ=0.90 reduces the mean unconditional wrong-commitment rate from 13.1% under chain-of-thought prompting to 8.9%, a 32.1% relative reduction, while increasing answered accuracy from 86.9% to 89.7% and answering 87.6% of questions. Both improvements hold for all eleven models and at every evaluated threshold. Critical-CoSQ and Adaptive-CoSQ provide neighboring operating points with 88.6% and 86.5% coverage, respectively, while remaining more reliable than the baseline. A secondary Natural Questions Short-Answer evaluation provides convergent open-form evidence. These findings show that self-assessment can support explicit, tunable answer-or-abstain decisions when an unsupported commitment is more costly than referral or review.

This work presents a proof of concept for auditable LLM-assisted medical billing verification based on a decision-basis contract. The contract separates deterministic checks of versioned fee-catalog rules from LLM-based assessment of free-text documentation. The deterministic layer resolves the applicable catalog release and checks code availability, quantity limits, and exclusions. The semantic layer classifies each claimed item as supported, contradicted, or missing required information. Support and contradiction require a verbatim evidence span; unavailable rule context, unsuccessful assessment, or missing required evidence prevents support through fail-closed abstention. We evaluated four locally run open-weight models on a synthetic catalog and 36 curated cases under the contract, an ablation without explicit documentation requirements, and an end-to-end baseline. Outcome agreement varied across models and showed no consistent advantage over the baseline. Explicit documentation requirements improved identification of missing information for all four models. The evidence gate also exposed cases in which correct raw judgments lacked valid evidence and were converted to incomplete decision-basis entries. The results show how explicit decision records can make rule findings, documentation judgments, and abstention reasons inspectable. Evaluation on real catalogs, independently annotated documentation, and with human reviewers is required to assess practical value.

Grounded question answering systems should answer only when the supplied evidence supports the answer. In multi-hop QA, this requirement is difficult because partial evidence can make an unsupported answer appear plausible. We study selective answering through evidence sufficiency boundaries: for the same question, a model should abstain under unsupported or partially supported context, answer when the context first becomes sufficient, and keep the answer stable when redundant evidence is added. We introduce Evidence Sufficiency Boundary Training, a generation-native training framework that constructs ordered evidence chains and supervises the abstain-to-answer transition directly. The method combines level supervision, a boundary flip margin, post-boundary stability, and answer recall protection. We build evidence chains from HotpotQA, 2WikiMultiHopQA, and MuSiQue, then evaluate models with chain metrics, raw QA utility, and unsupported-answer rates on external non-answerable sets. With Qwen2.5-3B-Instruct and LoRA adaptation, Evidence Sufficiency Boundary Training gives the strongest boundary localization among the tested systems, with flip accuracy of 0.807 compared with 0.781 for a token-level abstention baseline. It also achieves the lowest overall unsupported-answer rate on external non-answerable evaluation, 0.095 compared with 0.101 for the same baseline, while retaining competitive raw QA F1. The results show that grounded selective answering improves when training marks the evidence level where refusal should give way to answering.

Large Language Models (LLMs) show great potential as clinical agents, yet existing benchmarks reduce clinical workflows to static predictions or unconstrained Markov Decision Processes (MDPs) with coarse action sets. To address this, we introduce GPAgentBench-2K, the first Constrained MDP (CMDP) LLM-agent benchmark for primary-care clinical decision-making, constructed from expert-validated records of real-world GP encounters. Our environment models a full spectrum of six foundational clinical actions, imposes a topological workflow prior over the action space, and operationalizes safety-informed abstention as a first-class outcome. Evaluating 16 state-of-the-art LLMs reveals a significant performance degradation as the action space scales. Crucially, we uncover a clinical quality-safety gap: even frontier models with the highest diagnosis accuracy violate safety constraints in over half of high-risk cases. Finally, we establish a reference point using Constrained Group Relative Policy Optimization (C-GRPO), and show that while explicitly modeling constraints improves performance over unconstrained RL methods, it remains far from clinically acceptable safety.

KV-cache compression reduces LLM inference memory by evicting context tokens, but when the evicted tokens contain answer-bearing evidence, the model may hallucinate instead of recognizing that the compressed context is insufficient. We address this failure from a behavioral perspective: to our knowledge, this is the first work to formulate compression-aware abstention as a learning problem, in which a model learns to answer when supporting evidence survives compression and abstain when it does not. We construct supervision from compressor survival masks and tight answer-bearing spans, labeling examples as Confident when evidence survives and Abstain when it is removed. A 10.1M-parameter LoRA adapter trained on ~2.6K MuSiQue 2-hop QA examples reduces base-model hallucinations by 97% under prompt-style truncation while preserving correct answering on evidence-retaining examples. Unlike prompt-only abstention baselines, which over-abstain on many answerable high-retention examples, the trained adapter learns a conditional policy. We also evaluate the method under actual compressed-cache decoding, where multi-compressor training yields a 6-22x relative lift over the unaided base on evidence-retaining examples. Controlled-deletion experiments show that the learned behavior is driven by evidence content rather than input length alone.

Leak localization is usually evaluated as forced-choice prediction, although sparse hydraulic observations may not justify excavation. Here, we quantify a pressure-information limit and use it to recast localization as selective, evidence-gated decision-making. A physics-grounded executor falsifies competing leak, demand, sensor and valve hypotheses in a hydraulic twin. Deterministic code computes every number and every acceptance predicate; an independent large language model auditor may add a rejection but never overturn a failed check. Forced retrieval placed only 95 of 300 leaks in the correct zone. Across 550 mixed events, the gate acted on 223 (214 correct); on a third-party 33-leak benchmark, all four accepted events were correct. In a replay of 194 audited City D repairs, the pressure tier authorized five excavation recommendations, three matching the repaired district, while the district-inflow tier returned the correct district for 85 events. Observability limits with machine-checkable abstention enable auditable utility intervention.

When visual evidence is occluded or chaotic, models should abstain. In this paper, we show that Vision-Language Models (VLMs) can internally distinguish when abstention is required, but fail to express it anyway. We introduce TRAPSBench, a procedurally generated video benchmark of 1,404 matched physics pairs in which a single targeted change renders the outcome undeterminable from the visual evidence. Furthermore, we introduce Penalized Epistemic Calibration Score (PECS), a new robust metric that requires models to both answer correctly when the outcome is knowable, and abstain when the outcome is not. Across 16 VLMs spanning five families, spontaneous restraint is poor: the best PECS is 0.292. The bottleneck is expression, not perception: linear probes decode answerability from hidden states at up to 0.91 AUROC across physics domains; steering a single-layer void direction causally induces or suppresses abstention. Our results replicate across three open-weight families (Qwen, Gemma, LLaVA). The failure is also more pronounced in visual than textual uncertainty: models detect textual impossibility about 4x more readily than missing visual evidence. Closing this representation--output gap likely requires output-stage interventions.

Long-term agent memory is usually treated as select--store--retrieve, but retrieval does not decide whether contradictory, superseded, retracted, deleted, or stale records may support an outgoing claim. We introduce Governed Persistent Memory (GPM), an auditable bitemporal state-transition model with source-bound admission, derived lifecycle state, current public barriers, and fail-closed structured release. Five executable clauses cover ledger integrity, source binding, conflict isolation, non-revival after retraction or deletion, and exact claim closure over a fresh view at one verified head. On a prespecified hash-frozen 3,600-case GPM-ReleaseBench, GPM matches all complete outcomes; the strongest of three intentionally simple complete policies matches 1,800/3,600 and makes unmatched releases on 50% of violation cases. A separate sealed end-to-end service evaluation exercises real ingestion and release across eight query families. In its publicly disclosed V3 arm, the governed lane is correct on 2,400/2,400 clusters versus 600/2,400 for ungoverned local Qwen2.5-7B; it repairs all 1,800 baseline failures with no regression (one-sided 95% lower bounds 99.875% and 99.834%). A later V5 reseal over Chinese- and English-command arms, with generation-date pinning and no post-freeze reducer amendment, again obtains 2,400/2,400 per arm. A production-code-independent finite model explores 331,776 semantic and 1,990,656 query states without a full-contract counterexample, and a 100,000-trace three-engine differential yields zero mismatches. These are bounded contract and implementation results, not open-world model accuracy or evidence of world truth. Governed answers in the sealed service evaluation are deterministic service outputs; the 7B result is the ungoverned comparison, not a claim that a language model itself became perfectly accurate.

Tool-using LLM agents are commonly trained and evaluated in environments where tool calls succeed reliably, yet deployed tools can fail transiently, persistently, or silently. Robust recovery therefore requires more than repeated retries: an agent may need to retry the same path, switch to an alternative, or recognize that no viable path remains. We present BENCH2ROBUST, a framework that converts failure-free tool-use benchmarks into controlled stochastic environments with scenario-controlled solvability, where episodes explicitly require retrying, switching, or stopping after available paths are exhausted. We use BENCH2ROBUST to study two complementary interventions: structured runtime recovery context through Bayesian Tool Memory (BTM), and curriculum-controlled reinforcement learning. Across 7 models from 4 families and two multi-turn benchmark families, tool failures produce a near-universal robustness gap. On held-out Retail tasks, BTM improves robustness by up to 16.8 percentage points without retraining, while RL learns complementary recovery behavior that remains beneficial without inference-time BTM. Combining the two reaches 40.8-45.5% under injection while preserving failure-free performance. These results suggest that robust tool use benefits from combining environment-specific recovery knowledge with learned recovery behavior.

LLM analytics agents are evaluated on SQL syntax accuracy, but production failures look different: questions with two valid business definitions, questions the warehouse cannot answer, deprecated columns after a schema change, and queries that execute successfully while returning the wrong business number. No execution-match metric can score them. This paper introduces WarehouseReliabilityBench, 400 frozen tasks over two synthetic warehouses in which roughly half the correct responses are a clarification, an abstention or a refusal, with pinned denominators and a pre-registered paired bootstrap fixing each claim verb before the numbers existed. QueryProof, a 7B agent, uses rules derived from a semantic layer and physical catalog to determine its behaviour, and gates every answer on deterministic post-execution checks. On an 80-task synthetic test split evaluated once, QueryProof outperforms a direct-prompted 32B baseline by +0.237 [+0.112, +0.375] Business Truth Rate at 71.0% lower cost per correct answer; against a cost-matched few-shot baseline the accuracy gain holds but the cost difference does not resolve. This compares systems rather than model sizes: the 32B baseline receives none of the scaffolding. False success falls from 0.754 to 0.351 of returned answers, and no wrong number was returned on an answerable task (0 of 24), though 13 answers went to questions requiring clarification or abstention. Removing the routing layer changes little (0.562 against 0.537), so the result does not depend on escalation. Routing tuned on validation over-abstains on test, and the fitted confidence model loses to the heuristic it replaced. Resampling template families rather than tasks widens both accuracy intervals to include zero, so the effect's direction is better supported than its magnitude. The gain tracks the deterministic layer, though no component ablation was run.

Large reasoning models (LRMs) are prone to hallucination, which undermines their reliability and poses challenges for safe deployment. Hallucinations in LRMs arise from two distinct failure sources: reasoning hallucination, where flawed inference steps propagate to an incorrect conclusion, and knowledge hallucination, where the model lacks the requisite factual knowledge to answer the query. To address reasoning hallucination, we propose REIN, an alignment framework that trains LRMs to produce a structured reasoning sequence, \texttt{<think>} $$\rightarrow \texttt{<reflection>} $$\rightarrow $\texttt{<answer>}$, enabling explicit self-reflection before committing to a final answer. To address knowledge hallucination, REIN introduces a reward mechanism that encourages explicit abstention (e.g., "I don't know") when none of the sampled reasoning chains yields a correct answer, allowing the model to refrain from unsupported predictions. Extensive evaluations on mathematical and commonsense reasoning benchmarks show that REIN consistently improves selective accuracy, reduces incorrect-but-self-endorsed responses, and maintains high coverage compared with competitive baselines. Notably, REIN achieves these gains within a single forward pass, without requiring process supervision, inference-time controllers, external search, or multi-round critiques. Experiments on multiple backbones show that REIN reduces the hallucination proxy by $58\sim72\%$ relative to the base models while maintaining $86\sim91\%$ average coverage, and improves selective accuracy on attempted questions by $6.6\sim14.2\%$.

Modern models no longer keep a plain KV cache: latent caches, learned sparse selectors and recurrent states each carry the model's memory in a different form, and each fails differently under compression. We give a runtime observability contract that covers all four memory classes with three operators, instantiate it on six model configurations across five architecture families, and compose the per-stage bounds into an executable request-level risk ledger. Contracts carry their error metric as a type -- composition is only defined when metrics match, and this check rejected our own first composed chain; the repaired chain crosses metrics through two proved bridges, and whatever no formal system can certify is measured instead, dropping the composed tier to empirical automatically: every claim is certified, partially certified, or empirical, composition inherits the weakest tier, and the tier is decided by the machine. Replayed over $12.4$M entry reads and run under eight-way concurrency with per-request budgets and fail-closed identity attribution, the ledger quantifies the honest trade-off on today's witness and holds its risk budget with zero violations. A fused always-on probe observes a declared one-layer subset under CUDA graphs inside the serving noise floor. Applied to a served DeepSeek-V4 stack with a packed compressed-KV prototype, the same machinery localizes a silent corruption to a precise structural boundary -- exact in the eviction-free, identity-isolated regime, with every observed failure in an eviction or slot-reuse regime -- through a machine-adjudicated discrimination campaign whose calculus rejected two of our own confounded inferences along the way. All artifacts, guards, and the Lean development are released at https://github.com/metask-ai/witprobe-attention-memory; every number in this paper regenerates from the shipped artifacts by one command.

Interpretable classification often requires more than accurate predictions for real-life deployment: models should be transparent about the evidence behind their decisions and abstain when they cannot decide reliably. We introduce Fast Evidential Rule Learning (FERL), a method that learns interpretable, accurate fuzzy rule models whose outputs are evidential. Unlike post-hoc calibration, FERL's belief, plausibility, and abstention capabilities arise directly from the fuzzy memberships in a single deterministic pass, with no auxiliary head, held-out set, or repeated inference. Our theoretical analysis further shows that FERL is Lipschitz stable, which means that its evidential outputs vary smoothly with the input. Against state-of-the-art rule learners, FERL is statistically significantly more accurate across a 30 tabular-dataset benchmark ($+2.6\%$ average accuracy over the second best). Its native set predictions attain the best utility-discounted accuracy among credal classifiers ($u_{65}/u_{80}=0.80/0.83$ vs.\ $0.79/0.80$ for the naive credal classifier), at higher set coverage ($0.92$ vs.\ $\le0.82$). FERL also matches dedicated out-of-distribution detectors on tabular near-OOD detection ($77.7$ vs.\ $77.4$ AUROC for the strongest baseline). Under detector-class-disjoint concept-bottleneck evaluation, its it is within $2.3$ AUROC points of the strongest dedicated detector on both CUB and AwA2, while attaining the best AwA2 AUPR-Out ($68.3$) and novel-class rejection ($57.2$), while being able to name which attributes are anomalous.

Vision-language systems combine images with retrieved text, but these sources can disagree or jointly fail to support an answer. Reliable models must identify the trustworthy source and abstain when neither is adequate. Existing post-training objectives score instances independently and therefore do not enforce coherent behavior under counterfactual evidence changes. We introduce CARGO-VL, a group-relative framework that optimizes matched variants covering aligned, image-correct, text-correct, and both-wrong (A/V/T/N) evidence states as one bundle. Its objective couples condition-wise correctness with transition rewards for answer invariance, source equivariance, and answer-to-abstention switching, while a primal-dual controller balances unsafe answers against excessive deferral. We also contribute XMC (eXtended Modal Conflict), a four-condition conflict training resource, and evaluate transfer on CMC-Bench and Modality-Bias. Across multiple seeds, CARGO-VL improves conflict handling, unsupported-answer avoidance, and modality balance over pointwise baselines. Ablations identify complementary benefits from relational transition signals and adaptive risk control, supporting counterfactual consistency as a practical objective for reliable multimodal evidence arbitration.

Search-augmented large language model agents are increasingly capable of solving knowledge-intensive tasks, but their behavior when a multi-hop question is fundamentally unanswerable remains poorly understood. Existing abstention benchmarks largely expose defects at the surface of single-hop queries and therefore cannot reveal failures that emerge only after valid intermediate reasoning and retrieval. We introduce HopRefusalBench, the first controlled benchmark of refusal within multi-hop search, comprising 889 unanswerable questions constructed from KILT-grounded entity paths. It crosses three causes of unanswerability (answer unknown, false premise, and underspecified context) with root, middle, and terminal topologies, making premise verification, intermediate-bridge validation, and terminal stopping separately observable. We further propose a final-outcome taxonomy spanning target-aware refusal, pseudo-refusal, hallucinated completion, and search-budget exhaustion, together with source-aware trajectory metrics for post-trigger continuation and token waste. Across ten frontier proprietary and open-weight models in search-augmented mode, the best model achieves a target-aware correct halting rate (TCHR) of only 42.9%. Root and middle items are consistently harder than terminal items, and all models attain their highest TCHR on false premises and their lowest on underspecified questions. Yet when pooled across categories, 84.7--98.4% of each model's explicit refusal-like responses identify the correct rationale, localizing the main bottleneck to committing to an appropriate non-answer; failed trajectories instead diverge into hallucination or search-budget exhaustion. These results establish refusal in multi-hop search as a consequential evaluation problem and provide a foundation for diagnosing and improving the reliability of search-augmented agents.

Error-penalized scoring rules ($+1$ for a correct answer, $-λ$ for a wrong one, $0$ for abstaining) are increasingly prescribed against hallucination: a rational agent facing such a rule answers exactly when its correctness probability exceeds Chow's threshold $t^\ast=λ/(1+λ)$. We prove that a KL-anchored gradient learner can do the opposite. When abstention is a discrete action, the reward gradient and the anchor's restoring force are throttled by the same gate-saturation factor and die together: under explicit conditions (among them, blanket answering loses score in expectation and prompts share a bounded readout) the model drifts toward refusing everything, its mean training reward rising to zero like $1/t$ in training time $t$, so the curve reads as improvement while coverage collapses. The advantage estimator compounds the failure: in its sparse-answer regime, group normalization silently replaces every designed penalty with an effective penalty of one, moving the learned threshold from $t^\ast$ to $1/2$. The repair is structural: train a mandatory confidence report with a strictly proper score plus a correctness reward, and abstain only at deployment by thresholding the report. The always-emitted report has no gate to saturate, so no shared factor can kill its reward gradient and its anchor together, and its calibrated optimum is attracting. Simulations confirm every prediction, and experiments on language models at two scales confirm the mechanism live: the rule silences questions the models demonstrably still solve within ten optimizer steps, an ablation isolates the cause, and report-level training raises coverage, accuracy, and calibration together.

Model merging combines independently trained or fine-tuned models, but pairwise alignability does not imply globally consistent alignment. We formulate merging as a finite descent problem in which checkpoints are local objects, alignment maps are transitions, and cycle products are residuals. TwistedMerge is a conservative certification pipeline that separates fixed-chart averaging, synchronization-removable gauge inconsistency, a certified central obstruction on a specified comparison complex, and nonabelian holonomy. A residual is promoted to a cohomology class only after inverse-consistency, coefficient-identification, centrality, and closure tests; otherwise the method abstains and returns an ordinary or synchronized fallback. We prove a constant-edge no-go result, frozen-complex three-way and predeclared-family error-control theorems, and a refinement test for comparison-complex sensitivity. A planted neural alignment defect is removed by cycle-consistent synchronization, showing that a nonzero cycle score alone is not a higher obstruction. Controlled central systems recover the predicted non-coboundary and projective-rank behavior, while noisy estimates move from certification to abstention without false lifts on the tested controls. A trained low-rank-adapter audit shows that naive factor averaging depends on the chosen GLr representative, whereas global factor synchronization and dense-delta SVD are stable. On natural checkpoint collections, cycle residuals do not predict merge degradation and no natural central or period-index class is certified. The results position descent theory as a falsifiable certification and abstention framework.

Global LiDAR-to-BIM initialization must place a robot within an as-designed building model without a prior pose. In repetitive interiors, the principal failure mode is not low-confidence registration but confident aliasing: a submap can match several BIM regions with comparable scores, producing a high-scoring pose displaced by symmetry. We present Move First, Commit Later, a selective layer that treats a registration front-end as an evidence source and decides whether to commit. Candidates from multiple submaps are mapped to a common $\mathrm{SE}(2)$ anchor; a top-1 consensus $B_m$, invariant to non-champion multiplicity, aggregates cross-submap evidence; and topology serves only as a binary feasibility gate. The decision is typed--COMMIT, DEFER, or AMBIGUOUS($τ$), reporting the detected symmetry period--and reversible: symmetry-breaking motion upgrades AMBIGUOUS to COMMIT. On a real multi-room building and a controlled symmetric simulation, the layer commits correctly in every tested trajectory-scale configuration and otherwise abstains with a typed state, whereas forced-choice policies on the same front-end select wrong rooms in most cases. Committed anchors are within 0.02--0.36 m of an independent laser-tracker position reference and within $1.9^\circ$ of a BIM-registration orientation proxy. The evaluation covers one building and one front-end; the layer is designed to be front-end modular.

Recent advances in equipping Large Language Models (LLMs) with search tools and outcome-reward reinforcement learning (RL) have achieved new state-of-the-art results on open-domain QA tasks. However, we argue that current training paradigms harbor a critical vulnerability: they predominantly reward correct answers but fail to penalize fabricated ones when retrieval fails, thereby implicitly exacerbating hallucinations. To address this, we propose Abstention-Aware Reinforcement Learning (AWA-RL), which dynamically shapes the abstention reward utilizing the model's query-specific prior capabilities and continuous on-policy training observations. We also introduce a novel metric, RA-F1, to measure the capability-reliability trade-off. Compared to non-abstaining baselines, AWA-RL boosts absolute precision by up to 10.3% and overall RA-F1 by 2.9%, with only marginal sacrifice in raw accuracy. These results confirm that AWA-RL successfully yields highly capable and reliable search agents. The code, data, and model weights are publicly available at https://github.com/zfj1998/AWA-RL.

Agent systems based on large language models (LLMs) are increasingly deployed for autonomous tasks, yet existing evaluations mostly focus on task success rather than whether agents know when to abstain. This gap poses real risks: under ambiguity, conflicting constraints, or tool failures, agents may execute unintended and irreversible actions. To close this gap, we present the first systematic evaluation framework for agentic abstention: the calibrated ability of tool-using LLM agents to recognize when not to act. At its core, AgentAbstain is a paired-task benchmark built on an agent-native taxonomy of 8 abstention scenarios across pre-execution reasoning and runtime discovery. It contains 263 paired tasks across 42 executable sandbox environments, where each pair consists of a should-act task and a should-abstain variant produced through a controlled perturbation to the instruction, tool, or environment state. To scale this paired design and resist data contamination, we propose AbstainGen, a fully automated pipeline that synthesizes sandbox environments and generates paired tasks end-to-end, validated by deterministic replay and semantic LLM judges; fresh task instances can be regenerated on demand, and three independent annotators rate 94-98% of sampled tasks as well-designed. Across 17 frontier LLMs in 4 agent harnesses, the best agent (Gemini 3.1 Pro) achieves only 59.5% paired accuracy (correct on both the act and abstain sides of each paired task). More importantly, abstention capability is largely independent of general task-solving capability, indicating that scaling task-solving alone will not close this gap. We further identify failure modes such as post-hoc abstention, in which agents execute irreversible actions before recognizing abstention triggers. Our code and dataset are open-sourced at agentabstain.github.io.

Large language models (LLMs) are increasingly deployed in question answering (QA) systems, yet they may generate hallucinated or misaligned responses without reliable confidence estimates. Uncertainty quantification (UQ) offers a natural basis for selective answering, where a system answers only when its prediction is deemed reliable and abstains otherwise. However, existing uncertainty scores for LLMs are often heuristic: a threshold chosen on such scores does not, by itself, provide statistical guarantees on the error rate among accepted answers. We propose CIC, a confidence-interval-based calibration framework that converts arbitrary uncertainty scores into risk-controlled selective answering rules. Given a held-out calibration set, CIC evaluates each generated response using an application-specific alignment criterion and associates it with an uncertainty score and a binary error label. For each candidate uncertainty threshold, CIC estimates the acceptance-conditioned error rate and constructs a high-probability upper confidence bound using either Hoeffding-style or Clopper-Pearson confidence intervals. It then selects the largest threshold whose upper bound is below a user-specified risk level $α$, thereby maximizing the answering rate subject to a finite-sample reliability constraint. Under exchangeability, CIC guarantees with probability at least $1-δ$ that the selected threshold, if non-null, controls the error rate among accepted answers at level $α$. We evaluate CIC on both closed-ended and open-ended QA benchmarks across seven LLMs and multiple uncertainty estimators. Experimental results show that CIC consistently achieves valid risk control while retaining strong answering efficiency, providing a practical and statistically grounded mechanism for deploying LLMs in reliability-sensitive QA workflows.

In his 1996 doctoral thesis, Maurice Pagnucco created the first AGM-like abductive expansion operation. Taking his operation as a basis, as well as a taxonomy -- inspired by Atocha Aliseda -- responsible for highlighting and formalizing the main components of abductive reasoning, the main aim of this paper is to present a new paraconsistent AGM-like abductive expansion operation -- capable of assimilating contradictory explanatory hypotheses without trivialization and the consequent absurd epistemic state -- with its postulates and its transitively relational partial meet construction. To a large extent, the formal development presented in this paper was only made possible by the recent creation of the paraconsistent logic RCbr, an LFI (Logics of Formal Inconsistencies) that establishes properties especially relevant to belief revision contexts, in particular, the ability to be self-extensional -- i.e., to satisfy the replacement property. This is the first of two papers: the paraconsistent abductive expansion operation announced here -- which is part of a new system called AGMpabd -- despite bringing many interesting features, does not assign any relevant epistemic role to the paraconsistent operators of negation and consistency. Only in a second paper will an analogous paraconsistent abductive expansion operation -- which is part of another new system, AGMcircabd -- be enhanced in this direction. Nevertheless, to the best of my knowledge, the operation developed in this paper is the first of its kind in the AGM literature.

Consumer depth sensors such as the LiDAR scanner on recent iPhones provide metric range, but their useful range is short and their returns are sparse. We present DH-Active, a lightweight, training-free geometry back-end that treats the sensor as a metric ruler rather than the sole source of depth. Near-field returns anchor the metric relative pose of two views through PnP; visually trackable samples without a valid depth return are then triangulated under that pose. A parallax/reprojection gate abstains wherever the geometry is ill-conditioned, leaving an explicit hole and a selective score instead of forcing an estimate. The measured core front end, including spiral sampling, sparse back-projection, and hole taxonomy but excluding preprocessing and multi-view recovery, runs at 1.11 ms median latency on CPU (OpenCV using 14 threads), about 38 times faster than a DINOv2-L visual branch on GPU in our timing setup. Across two iPhone captures and the public TUM RGB-D and ARKitScenes benchmarks, held-out depth is recovered at 1.4 to 6.7 percent median relative error. In a controlled ARKitScenes protocol that uses only returns within 2 m to set scale and an independent laser scan as ground truth, DH-Active achieves 64.2 percent scene-median coverage of evaluable far-field candidates at 13.4 percent scene-median relative error; direct triangulation from the device trajectory is not usable. We also report the alternatives that failed in our tests: single-frame defocus, classical focus-stack depth, defocus-LiDAR fusion, point-to-point ICP over a good visual-inertial track, and attention-to-holes resampling. A 1.26 B learned model remains more accurate after oracle scale alignment. The contribution here is narrower: metric sparse depth, explicit abstention, zero learned parameters, and near-millisecond CPU cost.

Confidence is an estimate of the probability that a chosen answer is correct. Verbal confidence reports are widely used as uncertainty measures in large language models, but whether they are best understood as estimates of correctness is unclear. We test this with a two-stage abstention paradigm from the neuroscience of perceptual decision making: a model first answers and reports its confidence, then decides whether to commit it to a user or abstain. Across four non-reasoning models, prompt framings, and confidence formats, verbal confidence predicted the commit/abstain decision substantially better than whether the answer was correct. Calibrated token log-probabilities showed the opposite profile, with abstention-prediction coupled to correctness discrimination, the signature of an answer-evidence signal. After removing the variance verbal confidence shared with log-probabilities, the residual stayed aligned with commitment while its link to correctness fell to near chance. The dissociation generalised to four reasoning models across four benchmarks of varying difficulty, from hard multiple-choice to frontier-level freeform questions. Mechanistic analyses in Gemma 3 and 4 were convergent: a post-answer state known to causally support verbal-confidence generation already encoded the future abstention decision before the abstention prompt, organised mainly by that decision rather than by correctness, the two lying in approximately orthogonal directions in activation space. Steering along a verbal-confidence-specific direction causally shifted abstention. Verbal and log-probability confidence are thus not interchangeable: log-probabilities track answer evidence and correctness, whereas verbal confidence is better understood as a behaviour-facing readout of an internal commit-readiness state, challenging the practice of treating verbal reports as proxies for reliability.

We study the Bayesian fixed-budget best-arm identification problem in which a learner can abstain from making a terminal recommendation. Subject to an abstention budget $α$, we analyze the probability of undetected error--the risk of recommending a suboptimal arm without abstaining. Our central finding is that abstention induces a phase transition: without abstention, the error probability decays polynomially in the sampling budget $T$; in contrast, introducing any small positive abstention budget shifts this to an exponential decay. For Gaussian priors and rewards, in the regime $T\to\infty$ followed by $α\downarrow0$, we establish exact matching information-theoretic lower bounds and algorithmic upper bounds on the optimal error exponent, which takes the form $\exp(-\frac{α^{2}T}{8κ_ν^{2}})$. The hardness parameter $κ_ν$ represents the prior density of the top-two gap at zero, highlighting that nearly tied instances drive the fundamental error. We introduce an adaptive algorithm, PGWS, that successfully achieves this optimal exponent by expending its abstention budget on statistically ambiguous instances. We further demonstrate that this polynomial-to-exponential improvement is exclusively a Bayesian phenomenon--in the frequentist setting, abstention only affects lower-order exponent terms. We also extend our results beyond the Gaussian model.

LLM agents are expected to act over multiple turns, using search, browsing interfaces, and terminal tools to complete user goals. Yet not every goal is well specified or achievable in the available environment. In such cases, a reliable agent should recognize that further interaction is unlikely to help and abstain from additional tool calls. We define Agentic Abstention, the problem of deciding when an agent should stop acting under uncertainty. Unlike standard LLM abstention, which is usually evaluated as a single-turn answer-or-abstain decision, agentic abstention is a sequential decision problem: an agent can answer, abstain, or gather more information at each turn, and the need to abstain may only become clear after interacting with the environment. We study this problem across web shopping, terminal environments, and question answering, evaluating 13 LLM-as-agent systems and 2 agent scaffolds on more than 28,000 tasks. Our results show that the main challenge is not only whether agents can abstain, but also when they abstain. Some agents never abstain when they should, while others do so only after many unnecessary interactions. This gap is especially large on tasks where the instruction appears feasible until the environment reveals otherwise (e.g., no valid result matches the instruction). We further find that model scale, reasoning, and agent scaffolding affect abstention in different ways, where larger or more capable models sometimes perform worse at timely abstention. Finally, we introduce CONVOLVE, a context engineering method for improving agentic abstention that distills full interaction trajectories into reusable stopping rules. On WebShop, CONVOLVE substantially improves timely abstention without updating model parameters, raising Llama-3.3-70B's timely recall rate from 26.7 to 57.4. Our dataset and code are available at https://lhannnn.github.io/agentic-abstention

Autonomous AI systems are transitioning from advisory to autonomous roles for medication prescriptions. Recent United States bill H.R. 238 and Utah's prescription-renewal pilot both authorize AI to prescribe medications in an agentic capacity. While some regulatory guidelines suggest aggregate model performance metrics for clearance, they do not require i) calibrated per-prediction confidence for action-gated thresholds, ii) differentiated communication of uncertainty arising from model ignorance (epistemic) versus genuine clinical ambiguity (aleatoric), and iii) inferential transparency at the moment of decision that allows for liability allocation. Here, we present a regulatory and technical argument (tested with a survey of 136 U.S. prescribing clinicians) positioning these as minimum architectural requirements for safe autonomous prescribing. Our results suggest prescribing clinicians i) would not permit autonomous prescribing without a calibrated confidence-based escalation mechanism, ii) preferred a competing-options summary when uncertainty was aleatoric but shifted to abstention when uncertainty was epistemic, and iii) were only willing to accept additional liability when inferential transparency enabled a substantive judgment under acknowledged uncertainty. These findings indicate our recommended architectural features would encourage higher rates of clinician adoption, largely through collapsing much of what "autonomy" conventionally means. A system meeting these requirements would function less as an autonomous agent and more as a heavily supervised decision-support tool. As legislation and state pilots proceed, our technical argument backed by clinician perspectives provides opportunities for regulation to constrain the degree of autonomy ethically granted to AI in prescribing while aligning liability with the institutional actors who control system design and deployment.

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse tasks, yet they remain prone to generating hallucinations. Detecting these hallucinations is critical for deploying LLMs reliably in high-stakes applications. We present Grad Detect, a gradient-based approach for predicting hallucinations by analyzing layer-wise gradient patterns from a single forward-backward pass during inference. Our method shows that the internal gradient structure of a model carries rich information about the correctness of its output. This information is not accessible through output-level signals alone. We evaluate Grad Detect on several Q&A benchmarks across both hallucination detection and model abstention prediction, where it consistently outperforms confidence-based and sampling-based baselines. Through comprehensive layer ablation studies across all eleven models from four architectural families, we find that the final five layers concentrate over 97% of the discriminative gradient signal, enabling efficient deployment with minimal performance loss. Grad Detect provides a unified framework for predicting multiple dimensions of LLM reliability, offering strong predictive performance alongside interpretable insights into where and how model failures originate.

Forecasting leaderboards rank models by predictive quality, but their winners are often read as deployment-ready top-1 advice. That reading can fail when forecasts are passed through a fixed decision interface, such as an alert threshold, a top-k budget, or a switching-cost policy. We study when a forecast-side winner can be certified as deployment-actionable for a specified interface and deployed utility. We introduce a fail-closed certification protocol whose gates are sufficient evidential conditions for a strong claim: a friction-caused, non-tie, statistically supported, and recurrent deployment-side reversal. Traffic-Hourly provides a certified anchor: winners agree at zero friction, but positive switching friction makes the forecast winner deployed-suboptimal. A locked native audit tests overclaiming: across 22 verified candidates and 362 full-grid cells, 155 apparent forecast/deployment winner inversions are blocked before certification. The contribution is not a new forecaster, metric, or universal utility, but a conservative protocol for deciding when forecasting leaderboard winners should be read as deployment-actionable top-1 advice.

In large-scale enterprise settings, centralized multi-agent systems (MAS) are increasingly adopted, in which a coordinator delegates user requests to lightweight, domain-specialized sub-agents. While this architecture improves modularity, scalability, and cost efficiency, its reliability depends not only on accurate routing but also on sub-agents' ability to calibrate their responses to capability constraints. In particular, sub-agents built on smaller fine-tuned models often struggle with such calibration, leading them to over-answer ambiguous, underspecified, misrouted, or unsupported requests and produce hallucinated outputs instead of actionable feedback. To address this challenge, we present EARS (Explanatory Abstention for Reliable Sub-Agent Modeling), a production-oriented framework that reframes sub-agent abstention as an inter-agent communication protocol: a sub-agent does not merely abstain, but exposes an actionable failure state to the coordinator. EARS curates human-agent interaction data using an ensemble of calibrated LLM-as-a-Judge models, producing structured abstention labels and rationales under a taxonomy of sub-agent failure modes. These data are used to fine-tune sub-agents to detect failure conditions and return rationales for coordinator-level clarification, rerouting, or fallback. We evaluate EARS in a large-scale production e-commerce assistant supporting enterprise business intelligence workflows. EARS improves the overall response pass rate from 68.5% to 78.9%, demonstrating that sub-agent-side explanatory abstention improves MAS reliability.

Robotic assistance in household environments requires not only predicting where objects should be placed, but also reasoning about when objects should not be placed at all. Existing approaches to personalized object rearrangement primarily focus on placement decisions under the assumption of clean observations and complete actionability, limiting their applicability in realistic, cluttered, and partially erroneous settings. In this paper, we introduce APOLLO, a hybrid framework for abstention-aware personalized object rearrangement that combines a lightweight, personalized embedding model (PEM) with selective large language model (LLM) assistance. PEM is trained for each user-environment pair using a small number of demonstrations, operates entirely on CPU, and produces uncertainty estimates, which are used to selectively invoke LLM-based reasoning only for ambiguous decisions, balancing efficiency, privacy, and reasoning capability. To evaluate this formulation beyond existing benchmarks, we introduce APOR, a synthetic, LLM-generated dataset that captures room-level, multi-furniture environments, diverse organizational profiles, explicit abstention behavior, and noisy partial scene context. Extensive experiments on both PARSEC and APOR provide initial evidence that APOLLO improves over prior LLM-based baselines in controlled benchmark settings while substantially reducing LLM usage. Code is available at https://github.com/PaInt-Lab/APOLLO.

Large vision-language models (LVLMs) hallucinate: they assert visual details that the image does not support. A principled remedy is selective prediction with a distribution-free guarantee-verify each claim and abstain when the claim is not grounded, so that the hallucination rate among asserted claims is provably bounded. We show, however, that this guarantee is bought at a brutal price: to keep the hallucination rate below $5\%$ on a balanced object-existence benchmark, a state-of-the-art conformal filter must abstain on more than $80\%$ of claims. We argue that abstention is wasteful when more visual evidence is cheaply available, and introduce Budgeted Conformal Evidence Acquisition (BCEA), which replaces the binary answer/abstain decision with a three-way choice: answer, abstain, or acquire additional visual evidence by re-examining the image (zooming, cropping, or applying a claim-specific intervention) under a bounded compute budget. We make two observations. First, acquisition that is plugged naively into a calibrated filter breaks the statistical guarantee -- realized risk overshoots the target by up to $17$ points -- because the acquisition step destroys the exchangeability that conformal calibration relies on. Second, folding the entire acquisition policy into the score function and re-calibrating on post-acquisition scores \emph{restores} the finite-sample guarantee while still recovering coverage. BCEA further uses structured, claim-type-specific interventions. Across the POPE benchmark and COCO-constructed existence and spatial-relation claims, on four open VLMs, BCEA controls the hallucination rate at the target level and consistently improves coverage over a guaranteed-abstention baseline.

A constitution tells a language model what to value, but little tells us whether it does. Adherence is judged from outputs, and output evidence is most fragile on value conflicts, where what matters is not which value a model mentions but which one it is willing to sacrifice. We provide evidence that this arbitration can be read from activations in a structured margin readout. We introduce Constitutional Value Potentials (CVP). For each value we learn a scalar potential from the hidden state: an internal pressure to preserve that value, supervised not by the prompt but by an independent judge's verdict on which value the model's own response actually preserved. The signed difference of two potentials is a priority margin. A constitutional clause becomes the claim that a margin stays positive, and a single monitor score flags when it does not. The monitor predicts conflict violations with AUROC up to 0.95, beats a strong hidden-state probe, and generalizes to held-out synthetic conflicts across three Qwen2.5 scales. The signal appears as the answer begins, from the prompt tail and first response token. Read this early, the same signal reveals whether an adversarial priority hack has actually pushed the model toward a violation, rather than only whether the prompt looks adversarial. The same directions also support intervention tests: under selected steering settings, moving along a value direction shifts judged trade-offs in the intended direction. Together, these results suggest that some constitution-relevant priorities are accessible as activation-space margins, rather than only as output behavior.

Prompt-based LLMs are increasingly used for stance detection, but harder examples are not always repaired by clearer instructions, reasoning prompts, retrieval, or debate. We introduce SICI (Stance Inference Complexity Index), a seven-dimensional diagnostic measure of the semantic-pragmatic burden imposed by a target--text pair. Across SemEval-2016 and VAST, SICI predicts LLM accuracy better than surface proxies and shows substantial cross-scorer reliability ($α=0.771$). More importantly, LLM errors change regime as SICI increases: low-complexity examples invite over-attribution, especially Against predictions; intermediate examples form an unstable boundary; and high-complexity examples rapidly concentrate on None. This phase-transition-like structure persists across GPT-3.5, GPT-4o-mini, DeepSeek-V3, and GPT-4o, although stronger models move the boundaries. A 15-method intervention study further shows that prompting, retrieval, and debate often shift models along the attribution--abstention axis rather than removing the high-complexity bottleneck.

Hard-constraint decision systems usually veto infeasible candidates. This is too rigid when the system can act: if a known affordable repair would make an infeasible candidate feasible and valuable, rejection is a false veto rather than a ranking error. We introduce Q-RACL (Quantum Repair-Augmented Constraint Learning), a repair-before-veto framework that first defines RACL decision semantics and then identifies the single inference link where quantum feature access can be load-bearing. RACL accepts a candidate when a sequential repair plan restores feasibility and preference; otherwise it returns structured rejection credit. The hard link is repair-feasibility inference: which repair class restores feasibility from an observed candidate and context. We construct a discrete-logarithm-hidden RACL family where the repair class is a shifted interval rule in the latent exponent a = log_g(x), while the learner observes only x = g^a mod p. Under standard DLP-based learning separation, this coordinate is inaccessible to efficient raw-input classical policies but accessible to a quantum agent through Shor/Fourier structure. Across six primes and ten seeds, bounded raw-input classical policies and a wrong raw-Fourier encoding remain near chance, whereas the Q-DLP policy keeps false-veto rate below 1.1%, wins all paired seeds, and yields QNI_cond = 0.9777 to 0.9972. A classical DLog oracle matches it, isolating feature access rather than classifier capacity. Thus quantum AI is not added as a generic model upgrade; for this DLP-hidden repair family, it supplies the missing feature that closes the repair-before-veto loop.

Traditional TOPSIS derives its reference points -- the Positive Ideal Solution ($PIS$) and Negative Ideal Solution ($NIS$) -- from the observed alternative set, making rankings susceptible to misalignment with decision-maker (DM) requirements, sensitivity to outlier performances, and rank reversal. This paper proposes TOPSIS-RAD, which addresses these issues by incorporating two arrays of DM-defined reference levels. Vetoed Performance Levels ($VPL$) exclude non-viable alternatives before normalisation, preventing them from distorting the ranking frontiers. Desired Performance Levels ($DPL$) cap performances at the DM's desired level before normalisation, anchoring the $PIS$ in explicit aspirations rather than dataset extremes. Three toy examples demonstrate each mechanism: $VPL$ reshapes normalisation boundaries by removing a non-viable alternative; fixed $DPL$ frontiers stabilise rankings by limiting the influence of performances well above the desired level. The method preserves the familiar distance-based structure of TOPSIS while grounding the ranking in stable, DM-specified boundaries. Limitations and future research directions are also discussed.

The critical question after a correct driving veto is not only whether a maneuver is unsafe, but whether the blocked interaction admits a lawful, auditable, and responsibility-bounded repair. Prediction and game-theoretic planners can suggest plausible cooperation, yet they do not return a proof that the repair respects hard rules, right-of-way, cost allocation, and ego fallback. We introduce CARVE, Certified Affordable Repair of Vetoed maneuvers via Envelopes, a certificate architecture for prediction-free interactive repair. Given a vetoed maneuver, CARVE constructs a finite repair lattice and emits a structured certificate recording the binding rule, selected joint repair, right-of-way-scaled cooperation envelope, responsibility-weighted cost split, and ego-only fallback. This certificate view reveals the algorithmic bottleneck: multi-owner repair induces a product lattice $M = \prod_j |\mathcal{A}_j|$. We therefore introduce CARVE-Q, a verifier-shielded quantum-AI search layer that applies quantum minimum finding only to this black-box lattice while leaving all safety authority classical. In the conservative verifier-oracle model, exact classical minimum finding requires $Θ(M)$ queries in the worst case, whereas Durr-Hoyer/Grover minimum finding uses $O(\sqrt{M})$ oracle queries with high probability. We prove verifier-shielded certificate soundness, priority non-elicitation, black-box query separation, and finite-precision reversible-oracle constructibility. We then demonstrate state-vector minimum finding on CARVE repair oracles up to 65,536 assignments and validate certificate preservation on Lanelet2-grounded INTERACTION replay with 100% right-of-way respect, 100% blame consistency, and zero priority false positives. The result is a trust-bounded quantum-AI pattern for certified autonomy: quantum proposes; CARVE certifies.

Benchmarks for autonomous agents measure whether agents complete tasks, yet this framing is systematically blind to whether an agent should have proceeded at all. Agents trained under human-feedback objectives develop a structural tendency to proceed even when they lack the inputs, evidence, or authorization to act safely, a disposition we term compliance bias, because both the reward signal and the benchmark scoring regime treat proceeding as the correct default regardless of whether the preconditions for safe action are present. We make three contributions. We first show that compliance bias originates in reward hacking within human-feedback pipelines and is entrenched by prominent agent benchmarks, which either penalize agents for pausing or are architecturally unable to distinguish a principled pause from a silent failure. We then introduce a three-gap taxonomy of abstention-warranted scenarios, covering specification gaps where required information is absent, verification gaps where world state cannot be confirmed, and authority gaps where explicit authorization has not been given, which together provide a principled basis for constructing abstention-aware agent benchmarks. Finally, we propose abstention evaluation protocols (Safety Rate, Usability Rate, and Informed Refusal Rate) and report preliminary results across 144 enterprise agent scenarios and five model families, in which a runtime-enforced abstention mechanism achieves up to 89.2% hazardous-action blocking and 87.5% usability on authorized scenarios, demonstrating that the safety--usability tradeoff is tunable rather than inherent and that its shape varies substantially across model families. We treat this as preliminary work and offer the taxonomy and composite metrics as a starting point for further conversations.

Hard constraints are usually treated as terminal vetoes: once a candidate violates a requirement, the learned rule rejects it and any repair is handled outside the decision semantics. This misses a common deployed regime in which the system already knows a finite menu of modifications, such as adding a ticket option, changing a configuration, or requesting an available service upgrade. Existing constraint-learning, soft-relaxation, and recourse methods address nearby problems, but they do not learn whether an option should be repaired before being vetoed. We introduce Repair-Augmented Constraint Learning (RACL), a contextual decision framework that lifts known repair operators into the classifier semantics. A candidate is accepted when an affordable repair makes it feasible and preferred enough; otherwise the system returns a structured rejection credit and, when applicable, a repair plan. This repair-before-veto view strictly generalizes no-repair HASSLE-style semantics, reveals an irreducible false-veto gap for terminal-veto rules, separates binary-label non-identifiability from decision-rule learnability, and gives capacity and calibration bounds for the observed-feasibility shared-weight setting. Across controlled and DB1B-derived benchmarks, RACL recovers the intended credit and repair structure. On the hardest raw-data-derived tier, validation-selected RACL reduces false vetoes to 10/4039 (FVR 0.0025), versus about 1064/4039 for the strongest repair-search black-box baseline, while making the FVR/EDR trade-off explicit.

Interactive driving exposes a failure mode that is easy to miss in rule-aware autonomous-driving stacks: a hard-rule margin can be negative for an ego candidate even though a small lawful accommodation by a non-priority agent would restore feasibility. Existing rulebooks, shields, and reachability filters are strong at vetoing unsafe actions, while prediction-based planners model likely responses. Neither returns a runtime proof object that states which bounded multi-agent edit repairs the maneuver, who owns the edit, whether the request is right-of-way affordable, and what ego fallback remains if the request is not observed. We formulate this missing object as interactive repair certification and introduce CARVE, a prediction-free certificate layer over a finite lattice of ego-owned and agent-owned tactical operators. Agent-owned requests are admissible only inside $B_j(s) = β(π_j)α_j^{\max}(s)$, a cooperation envelope that separates kinematic reachability from normative priority. The resulting certificate records the binding rule, repair category, repair set, responsibility-weighted cost split, and fallback. On 589 Lanelet2-geometry-grounded INTERACTION replay episodes, CARVE-Greedy accepts 98.64% of initially vetoed maneuvers and recovers 370/378 human-resolved false vetoes, while preserving 589/589 right-of-way respect, zero priority-agent false positives, and 400/400 negative-stress vetoes. We prove certificate soundness, structural right-of-way respect, exact finite-lattice minimality, fallback contingency, and blame-consistency conditions. CARVE does not predict or require another driver's compliance; it certifies whether a proposed interaction is bounded, attributable, and normatively admissible under declared assumptions.

Graph classification is a core task in graph data mining with widespread real-world applications. Recent advances in graph neural networks (GNNs) have led to substantial performance improvements for graph classification. However, existing GNNs are typically forced to make predictions even under high uncertainty or unknown conditions, resulting in unreliable decisions that can severely impact downstream tasks, particularly in safety-critical scenarios. To address this critical limitation, we propose AbstainGNN, a novel and theory-driven framework for graph classification with abstention, which enables GNNs to reject uncertain predictions instead of producing incorrect decisions. Specifically, AbstainGNN explicitly models both the predictive function and the abstention function, allowing for effective utilization of graph structural information. Moreover, unlike existing heuristic abstention methods, we theoretically characterize the trade-off between classification errors and rejection costs from a PAC-Bayesian generalization perspective, and derive a unified learning objective for model optimization. Guided by this theoretical insight, we further develop an efficient two-stage training strategy consisting of predictive function warm-start and abstention function calibration. Extensive experiments on five benchmark datasets show that AbstainGNN outperforms existing abstention methods, achieving superior classification performance under the same rejection rates.

Test-time scaling for vision-language-action (VLA) policies, methods such as RoboMonkey, SEAL, MG-Select, and V-GPS, samples K candidate action chunks at inference and executes the verifier-best. When all K candidates are unsafe, the system executes a violating action with no warning. We propose BOKBO, the first conformal abstention layer for K-sample VLA inference, providing finite-sample distribution-free guarantees on executed-violation rate. We provide both global and per-task (Mondrian) variants, with the per-task variant closing the conditional gap on the hardest tasks. Our analysis exposes a structural failure of policy-internal nonconformity scores under perturbation-based K-sampling: the base-policy confidence proxy and K-sample disagreement correlate at 0.98 with the action-noise hyperparameter $σ$, while correlating at the noise floor with actual safety violations. We test the failure's scope by replicating the analysis under token-level temperature sampling and find the failure is mechanism-specific and partially mitigated under policy-stochasticity-based sampling. A learned violation predictor conditioned on semantic visual features and task identity supports tight calibration: at $ε$ = 0.05 on libero_object_temp_x0.1 with OpenVLA-OFT, the conditional CRC bound holds on 86% of bootstrap splits with 78% coverage and 70% net task success. Mondrian-BOKBO raises the minimum per-task conditional hold fraction from 0.71 to 0.93. Results are stable across 5 training seeds, replicate within bootstrap noise on $π_0$-FAST, hold on libero_spatial_temp_x0.1 as a co-equal benchmark, and survive four within-suite distribution shifts. We additionally identify and correct a methodological pitfall: globally-set force thresholds well below expert-typical manipulation forces conflate unsafe behavior with normal manipulation, inflating violation rates by $5\times$.

Production AI systems often operate with incomplete, conflicting, or insufficient evidence. Forced classifiers collapse such cases into action labels, while generative systems can produce outputs that are difficult to interpret as auditable execution decisions. We study operational decision control for AI systems, where uncertainty must be explicitly routable, policy-governed, and auditable rather than hidden inside forced predictions or free-form generation. We present EvaluatorDPT, a bounded decision-control model that predicts YES, NO, or TBD, where TBD is learned as a deferral outcome rather than added only as a post-hoc confidence rule. The model uses a transformer encoder with a primary bounded-decision head and structured auxiliary channels for values and emotions/sentiments. The interface is domain-agnostic in form: a deployment domain supplies evidence and policy thresholds, while the model emits a bounded distribution that can be controlled at inference time through recorded operating thresholds and, when validated, auxiliary semantic signals. For the evaluated model version, we report decision performance on held-out validation and test splits; auxiliary emotion metrics are omitted because the emotion head is disabled for this evaluation. On the held-out test split (n=44,597), the model achieves Accuracy = 0.8260 and Macro F1 = 0.8252, with per-class F1 of 0.8314 (YES), 0.8486 (NO), and 0.7956 (TBD). The evaluation record also includes calibration evidence (ECE = 0.0338 on validation), threshold-sweep outputs, multi-seed stability checks, confusion matrices, and reproducibility commands. Our main contribution is a bounded execution interface in which deferral is learned, inference-time routing remains inspectable, auxiliary signals provide a path to auditable behavior control, and evaluation evidence supports external review.

Large language models (LLMs) define a distribution over text, which can be viewed as a probabilistic representation of uncertainty: sampling K responses yields a belief state - responses a model deems plausible. Existing work exploits this representation for narrow tasks like either decoding or selective prediction, and often requires manual interventions, not controlling generation directly. We propose Belief-Augmented Generation (BAG): grounding LLMs in their own belief state via the prompt and letting them reason over these K samples to decide on and execute a conversational strategy: clarify, abstain, or answer. In a multi-turn ambiguous question answering (QA) setting, we find that LLMs by default rarely clarify or abstain, ignoring uncertainty about the input (aleatoric) or facts (epistemic). BAG improves QA accuracy across six models and yields strategy decisions more faithful to their belief state than prompt-only baselines. Disentangling when to clarify from when to abstain, however, remains challenging.

Validating an LLM judge against human annotations usually means reporting several agreement statistics: accuracy, precision, recall, $F_1$, Cohen's $κ$, and one or more rank correlations. A survey of 24 recent LLM-as-judge papers finds metric choice entangled with the judgment scale, tie handling, invalid outputs, and abstention handling, and those choices rarely stated. For binary criteria -- the common case in rubric-based evaluation, where each criterion is graded MET or UNMET -- most of the reported numbers are redundant: Pearson's $r$, Spearman's $ρ$, Kendall's $τ_b$, the phi coefficient $φ$, and the Matthews Correlation Coefficient all reduce to a single number on non-degenerate binary data, so reporting several of them only creates an illusion of corroborating evidence. Cohen's $κ$ is the one agreement coefficient that adds information: it shares $φ$'s numerator but normalizes differently, and the gap between them measures how far the judge's positive-label rate has drifted from the human's. We then trace what changes when a judge may abstain with a CANNOT_ASSESS verdict: the three common ways of handling abstentions are not interchangeable preprocessing choices but answer different questions, and they break the binary equivalences. The same equivalences reappear, up to a negligible finite-sample correction, for multi-judge ensembles scored with Fleiss' $κ$ or Krippendorff's $α$. We close with a reporting checklist that names the judgment scale, the abstention and tie handling mode, coverage, the confusion matrix, and the aggregation level alongside any scalar agreement coefficient.

Large language models often generate confident but incorrect answers rather than abstaining when uncertain. This problem is particularly acute for small language models (SLMs), where computational constraints and autonomous operation amplify the need for reliable uncertainty detection. We propose Second Guess, a lightweight, parameter-free prompting technique for abstention in multiple-choice question answering (MCQA) that is well-suited for SLMs. Our key empirical insight is that models which truly know an answer will select it consistently, while uncertain models exhibit unstable behavior when an ``I don't know'' option is added. Evaluated on four open models (2B-8B parameters) and four benchmarks, Second Guess achieves the highest composite risk improvement of 10.81%. Notably, it maintains an 8% composite risk improvement on fine-tuned models where entropy-based methods degrade, and improves most for lower-performing models. All code and results required to reproduce this work is available in https://github.com/Mystic-Slice/second-guess

An agent must act on the situation before it, learn what it cannot yet represent, and model other agents well enough to coordinate. These faculties are usually realized by separate mechanisms, yet they share a failure mode: the situation can exceed what the agent can currently represent, and the honest response is then a principled refusal that says what was missing. We develop a small cognitive architecture in which these limits arise from a single quantity. An Interpretation-Decision Unit (IDU) interprets a content vector through a family of regimes - local representational frames with private bases - and decides which actions it licenses; a scalar residual of the content against the active regimes' representational scope drives the unit. Low residual with a clean licensing emits an action; otherwise the unit re-interprets, attempts a description-length-justified expansion, or halts with a typed, witnessed terminal. We prove the unit is total and deterministic: for any content and fixed configuration it halts in finitely many bounded-cost steps with a unique terminal witness, so abstention carries its cause by construction. By binding the architecture's open parameters without changing its mechanics, the same residual-against-scope constraint recovers three documented phenomena at three scopes: the typology of not-knowing (typed abstention); a forced misunderstanding between agents, localized to one shared concept and invisible to the agent committing it (bounded empathy); and prerequisite dependence in learning derived from a bounded focus window rather than posited (developmental prerequisites). Each instantiation is worked for a natural and an artificial agent and states a falsifiable prediction, so one constraint can model limits in both human and machine cognition. The account contributes a unification and a notion of accountable abstention, typed and witnessed by construction.

Vision-language models (VLMs) are used as high-level planners for embodied agents, translating natural language instructions and visual observations into action plans. While prior work has studied abstention in LLMs, existing benchmarks are largely text-only and do not capture the perceptual grounding and physical constraints inherent to embodied robotics environments. In such settings, abstention requires recognizing when instructions are ambiguous, physically infeasible, based on false premises, or otherwise unresolvable given the available sensory modalities and context. To address this gap, we introduce a taxonomy to categorize abstention in the context of embodied robotics and present RoboAbstention, a scalable and auditable framework for generating abstention instructions grounded in images gathered from five robotics datasets. RoboAbstention instantiates the taxonomy through a three-phase pipeline: (1) structured visual grounding, (2) deterministic constraint derivation, and (3) controlled instruction generation via category-specific templates. This enables the construction of a diverse dataset with verifiable abstention conditions. We evaluate several frontier VLMs and find that all models exhibit significant weaknesses in abstention, including those with advanced reasoning capabilities. The best-performing model, Gemini 2.5 Flash, abstains on only 39.0% of our 6,069 benchmark instructions, while the embodied planner Gemini Robotics ER 1.6 Preview abstains on just 16.5%. We further explore methods for improving abstention in VLM planners, such as defensive prompting and in-context learning, and find that these interventions substantially improve performance, reaching 93.6% abstention rate for Gemini Robotics ER 1.6 Preview and 88.6% for GPT 5.4 Mini, yet no approach fully solves the problem. We open-source RoboAbstention at https://purseclab.github.io/RoboAbstention/.

Self-supervised learning (SSL) is now a standard way to pretrain medical image models, but performance is still mostly judged by downstream accuracy. For safety-critical screening tasks such as diabetic retinopathy grading, this is not enough: a model must also know when its predictions are unreliable and defer uncertain cases for clinical review. In this work, we examine how the length of SSL pretraining influences calibrated confidence and confidence-based abstention. We evaluate multiple SSL checkpoints under a fixed fine-tuning protocol and assess calibrated confidence, coverage, selective accuracy, and selective macro-F1. Across datasets and data regimes, SSL pretraining improves selective prediction compared to training from scratch. Unlike prior SSL studies that primarily evaluate downstream accuracy or AUROC, we analyze how SSL pretraining duration influences confidence behavior under calibrated confidence-based abstention. However, once accuracy saturates, selective performance can still change markedly across checkpoints, and longer pretraining does not consistently improve reliability. These results underscore the importance of abstention-aware evaluation and suggest that pretraining length should be treated as an important reliability-related design choice rather than only a computational detail. Code is available at GitHub.

We study group decision-making in artificial societies where the rules of play are themselves subject to collective amendment. Using the self-amending game Nomic, we compare multiple scales across two LLM families and find that collective adaptation does not improve monotonically with model size. Instead, both families exhibit a narrow mid-scale regime that supports sustained rule adoption, diverse amendments, and balanced consensus. Smaller models tend to remain rule-inert, whereas larger models often converge on restrictive voting patterns, and heterogeneous mixed-size groups collapse into veto-driven gridlock. These cross-scale contrasts persist under temperature perturbations and under a shift from unanimity to majority voting, although latent-state structure varies by family and scale. Hidden-state divergence alone does not explain collective performance: high representational divergence can coincide with poor behavioural outcomes. Linear probes reveal regime-selective coupling between latent vote-predictive signals and collective behaviour, but decodability is necessary rather than sufficient for adaptive play. Overall, the recurring regularity is non-monotonicity, not the particular scale at which the optimum appears. Self-amending games therefore provide a controlled testbed for studying collective adaptation in artificial societies beyond raw model scale.

Large language models (LLMs) have revolutionized automated code generation. One serious concern, however, is the so-called ``hallucination'', i.e., LLMs may generate seemingly plausible but functionally incorrect code. In this paper, we study the task abstention problem, i.e., determining whether a given LLM should abstain from performing a specific code generation task to avoid likely hallucination. Our approach features a calibrated abstention rule, grounded in the principles of multiple hypothesis testing. The rule assesses generation consistency through code execution outcomes, allowing it to handle syntactic diversity of semantically equivalent code without reliance on oracle test cases or external databases. We prove that our approach provides a rigorous, distribution-free theoretical guarantee on its abstention decisions. We evaluate our method on benchmark datasets using several open-source code LLMs. Results show that our method allows generative models to more accurately and efficiently identify and abstain from tasks that induce hallucination compared to existing techniques, providing a reliable mechanism for safer and more robust code generation.

Fairness in toxicity classification involves three integrated axes: ranking, calibration, and abstention. Training-time interventions and post-hoc safety mechanisms cannot be evaluated independently because the former determines the efficacy of the latter. We compare Empirical Risk Minimization (ERM), instance-level reweighting, and Group DRO across these axes, combined with temperature scaling, confidence-based abstention, and per-identity threshold optimization. Evaluation uses subgroup AUC, BPSN/BNSP AUC, error gaps, and per-subgroup Expected Calibration Error (ECE) with bootstrap CIs ($n = 1000$). We report four findings. (1) Calibration disparity is a hidden fairness violation. ERM has near-perfect aggregate calibration ($0.013$) but is significantly miscalibrated across all identity subgroups ($+0.029$ to $+0.134$). (2) Training interventions reshape rather than eliminate disparity. Reweighted ERM improves ranking (BPSN AUC $+0.06$ to $+0.12$) but worsens the calibration-fairness gap by up to $+0.232$. Group DRO eliminates calibration disparity but only by becoming uniformly miscalibrated globally (ECE $0.118$). (3) Post-hoc methods inherit training failure modes. Temperature scaling fails because miscalibration is non-uniform. Confidence-based abstention works under ERM but breaks under DRO, where the risk-coverage curve rises with deferral. (4) Abstention itself is unfair. Confidence-based deferral helps background content far more than identity-mentioning content. We argue that SRAI fairness requires a multi-axis framework: methods that differ only in aggregate ranking can differ sharply in failure modes that determine real-world harm.

Retrieval-augmented generation (RAG) improves large language models (LLMs) by incorporating external evidence, but it also introduces knowledge conflicts when retrieved contextual knowledge (CK) and parametric knowledge (PK) disagree or are both unreliable. Existing approaches mainly coordinate which source to use, without explicitly asking whether each answer path is correct. We argue that faithful RAG requires LLM self-awareness, namely the ability to recognize the limits of its own knowledge and reasoning. To ground this problem, we construct a model-specific, ground-truth-aligned knowledge-conflict benchmark by evaluating LLM backbones on PK-only and CK-conditioned answer paths over approximately 69K query-context instances per backbone, drawn from five conflict-QA datasets. We then introduce SABER, a Self-Aware Belief Estimator for RAG that requires no LLM fine-tuning. SABER combines a self-prior with PK-side and CK-side conditional reasoning representations from multi-trace inference, then estimates reliability beliefs with two lightweight predictors to drive a 4-cell decision over trust PK, trust CK, trust either, or abstain. Across four LLM backbones, SABER improves end-to-end accuracy and conflict-specific faithfulness over ten inference-time and fine-tuning baselines, with the largest gains on conflict-heavy datasets. Under abstention, SABER's risk-coverage curve Pareto-dominates every prompt-based abstainer, providing a tunable balance between coverage and answer risk. Our code is available at https://github.com/xizhu1022/SABER.

Standard selective prediction methods typically estimate uncertainty from the output of a single predictive branch. While effective for general uncertainty estimation, these approaches often struggle under partial observability, where local temporal evidence can be contradictory and standard confidence scores become misleading. We introduce CHASE (Competing Hypotheses for Ambiguity-Aware Selective Prediction), a selective prediction framework that explicitly compares structured temporal explanations to determine whether to commit to a decision or abstain. Because genuine ambiguity causes the score gap between competing hypotheses to collapse, CHASE optimizes a ranking-aware selector over these hypothesis margins to globally separate safe commitments from fundamentally uncertain ones. We evaluate this framework on the problem of hidden connectivity inference, utilizing a controlled, physically grounded simulator inspired by the dynamics of giant unilamellar vesicles (GUVs), alongside zero-shot qualitative transfer (without retraining or fine tuning) to representative real GUV videos. Our experiments demonstrate that explicitly reasoning over competing hypotheses provides a superior balance of metrics. Compared to canonical uncertainty baselines, CHASE achieves statistically significant gains in overall no-abstain accuracy, three-way accuracy, and overall ambiguity-aligned abstention (at 80% coverage). Specifically, it yields up to an 11.0% relative mean improvement in overall alignment, alongside up to an 8.8% relative boost in three-way accuracy in the very-high ambiguity regime. By maintaining a selective risk boundary strictly at par with the best baselines at 80% coverage, and reducing overall risk by 9.9% at 90% coverage, this framework offers a more reliable approach to decision-making under structured ambiguity.

When language models lack relevant knowledge for a given query, they frequently generate plausible responses that can be hallucinations, rather than admitting being agnostic about the answer. Retraining models to reward admitting ignorance can lead to overly conservative behaviors and poor generalization due to scarce evaluation benchmarks. We propose a post hoc framework, Conformal Abstention (CA), adapted from conformal prediction (CP) to determine whether to abstain from answering a query. CA provides finite-sample guarantees on both the probability of participation (i.e., not abstaining) and the probability that the generated response is correct. Importantly, the abstention decision relies on prediction confidence rather than the non-conformity scores used in CP, which are intractable for open-ended generation. To better align prediction confidence with the model's ignorance, we introduce a calibration strategy using representation geometry within the model to measure knowledge involvement in shaping the response. Experiments demonstrate that we improve selective answering significantly with 75 percent conditional correctness.

LLMs utilizing chain-of-thought reasoning often waste substantial compute by producing long, incorrect responses. Abstention can mitigate this by withholding outputs unlikely to be correct. While most abstention methods decide to withhold outputs before or after generation, dynamic mid-generation abstention considers early termination of unpromising reasoning traces at each token position. Prior work has explored empirical variants of this idea, but principled guidance for the abstention rule remains lacking. We present a formal analysis of dynamic abstention for LLMs, modeling abstention as an explicit action within a regularized reinforcement learning framework. An abstention reward parameter controls the trade-off between compute and information. We show that abstaining when the value function falls below this reward strictly outperforms natural baselines under general conditions. We further derive a principled and efficient method to approximate the value function. Empirical results on mathematical reasoning and toxicity avoidance tasks support our theory and demonstrate improved selective accuracy over existing methods.

Reinforcement fine-tuning improves the reasoning ability of large language models, but it can also encourage them to answer unanswerable queries by guessing or hallucinating missing information. Existing abstention methods either train models to produce generic refusals or encourage follow-up clarifications without verifying whether those clarifications identify the key missing information. We study queries that are clear in meaning but cannot be reliably resolved from the given information, and argue that a reliable model should not only abstain, but also explain what is missing. We propose a clarification-aware RLVR reward that, while rewarding correct answers on answerable queries, jointly optimizes explicit abstention and semantically aligned post-refusal clarification on unanswerable queries. Using this reward, we train Abstain-R1, a 3B model that improves abstention and clarification on unanswerable queries while preserving strong performance on answerable ones. Experiments on Abstain-Test, Abstain-QA, and SelfAware show that Abstain-R1 substantially improves over its base model and achieves unanswerable-query behavior competitive with larger systems including DeepSeek-R1, suggesting that calibrated abstention and clarification can be learned through verifiable rewards rather than emerging from scale alone.

Effective abstention (EA), recognizing evidence insufficiency and refraining from answering, is critical for reliable multimodal systems. Yet existing evaluation paradigms for vision-language models (VLMs) and multi-agent systems (MAS) assume answerability, pushing models to always respond. Abstention has been studied in text-only settings but remains underexplored multimodally; current benchmarks either ignore unanswerability or rely on coarse methods that miss realistic failure modes. We introduce MM-AQA, a benchmark that constructs unanswerable instances from answerable ones via transformations along two axes: visual modality dependency and evidence sufficiency. Evaluating three frontier VLMs spanning closed and open-source models and two MAS architectures across 2079 samples, we find: (1) under standard prompting, VLMs rarely abstain; even simple confidence baselines outperform this setup, (2) MAS improves abstention but introduces an accuracy-abstention trade-off, (3) sequential designs match or exceed iterative variants, suggesting the bottleneck is miscalibration rather than reasoning depth, and (4) models abstain when image or text evidence is absent, but attempt reconciliation with degraded or contradictory evidence. Effective multimodal abstention requires abstention-aware training rather than better prompting or more agents.