Runtime Enforcement for AI Agents

In 2026, cybersecurity evaluations involving OpenAI, Anthropic, and Google agents reached real systems outside their authorized test scope. The paths were different. OpenAI agents exploited research infrastructure, coordinated across runs, and compromised parts of Hugging Face's production environment. Anthropic reported cases in which a misconfigured third-party environment exposed real systems to agents pursuing simulated cyber tasks. In a separately reported evaluation, Google's Gemini accessed three real organizations through an unintended internet route; Google stated that the model stopped in all three instances. Taken together, the cases show why an evaluation cannot rely on an assumed boundary. That boundary must be verified while the agent is operating. This comparative instrumental case study develops a Proactive Agent Security Assurance Cycle (PASAC) and a five-layer Boundary Assurance Stack. The framework combines risk-tiered task design, executable scope contracts, pre-run validation, least-capability access, independent egress enforcement, credential restrictions, cross-run monitoring, automatic stop conditions, and evidence-based reauthorization. A leading-indicator model, nine design propositions, and seven falsifiable hypotheses turn these lessons into a testable research program. Because the public Gemini record is limited to attributed statements and journalism, its detailed causal mechanism remains provisional. The central conclusion is straightforward: proactive agent security requires continuous assurance across the full execution system, not confidence in any single sandbox or safeguard.

Agents are deployed in applications from trip planners and stock trading to IT incident triage. In most cases, LLM agents work autonomously with minimal rule-based safeguarding, leading to cost and safety issues from irreversible actions. Recent works resolve this either by using a safeguard agent to monitor behavior or evaluating logs post-hoc. The first adds cost and latency to every step; the second delivers its verdict after the run, when tokens are burned and damage is done. To overcome this, we propose OnTrack, a streaming monitoring mechanism that compares an agent's steps and dependencies against recorded successful runs to alert users or block the agent in about a millisecond per step. We study this problem in three regimes of decreasing access: full reference access (historical runs and tool schemas), intermediate access (only tool schemas), and no prior knowledge (only step logs as generated). Expectation of OnTrack's monitoring capabilities reduces as data access drops, ranging from plan violation detection to identifying loops, stalls, and repeated tool calls. Finally, we evaluate OnTrack using SWE-bench trajectories. Based on the first 8 steps, our method ranks failing trajectories below succeeding ones better than content similarity approaches (+0.057 AUROC). With an abort policy, we save about 18% of compute that would be burned on failing runs, where 83% of interrupted runs were actually heading to failure (5 out of 6 aborts were correct).

Large language model agents can invoke tools fluently, but enterprise workflows demand more than selecting the right tools: actions must strictly comply with organizational policies, tool feedback often conceals hidden side effects under partial observability, and long-horizon tasks require persistent state tracking across multiple records. To address these challenges, we introduce E-Ledger, a multi-agent harness for safe and persistent execution. E-Ledger employs a code approval layer that checks every proposed action against policy before execution, and maintains a world ledger of verified hidden rules alongside evidence-backed dynamic state. Because hidden rules are typically unknown a priori, we further propose WorldAbduct, an abductive, world-model-driven harness evolution framework. WorldAbduct diagnoses execution trajectories across four complementary views (state consistency, world-observation gap, policy-gate correctness, and goal judgment) to hypothesize latent rules, and verifies them through targeted abductive interactions before integrating them into the ledger. On the enterprise benchmark World of Workflows, E-Ledger with WorldAbduct improves safe task completion across four LLM backbones, outperforming the strongest evolution baseline by 5--15 percentage points. Experiments in ScienceWorld and DiscoveryWorld further show that abductive harness evolution carries over to scientific environments. Our code is available at https://github.com/HKUST-KnowComp/E-LEDGER-WorldAbduct.

Tool-using LLM agents violate the policies they are deployed to enforce, often silently. Prior defenses hand-write rules, query an LLM verifier per action, or compile policies through heavyweight formal machinery. Naive compilation fails: extracted rules block the tool satisfying their own precondition, or read arguments their tool lacks. NOMOS, a four-pass compiler, turns a natural-language policy into a deterministic tool-call gate; static verification with tool-schema-level checks alone (no prover, solver, or LLM) repairs or rejects 37% (airline) and 13% (retail) of candidates, without which most shipped rules are inoperable. Replaying compiled rules over undefended transcripts flags bindings that refuse legitimate work (a development binding refused 95.9% of task-passing calls); no evaluation binding is flagged. On $τ^2$-bench the gate cuts violations of reference-encoded clauses among state-changing calls from 66.3% to 2.6% (airline) and 30.8% to 6.9% (retail), raising airline task success significantly for $2 \le k \le 4$; a 26B on-premise compilation is not significantly worse than hand-written or frontier-compiled rules. Unlike AgentDojo's shipped defenses, it reaches a zero attack success rate (ASR) on banking, where nine attack families collapse onto three structural rules. On the other three suites its ASR is at most 3.6%, from goals with no tool call to govern and one write admitted by a binding weaker than its clause; a second agent model, Llama-3.3-70B, reproduces the effect on both benchmarks. Decisions take microseconds without an LLM call, at a domain-dependent benign-utility cost; compilation runs on-premise on open-weight gemma-4-26B.

Coding agents increasingly share a workstation while drawing on separate providers and subscription allowances. A second agent can inspect a completed answer, but the call spends another pool and may provide no substantive finding. We describe an advisory cross-provider review contract: distinct resource pools, bounded execution, restricted reviewer capabilities, complete input delivery, usable semantic output, explicit failure states and durable per-attempt evidence. In a controlled, agent-authored pilot of 20 paired development turns, eight had a material reviewer finding (95% exact interval 19.1-63.9%). A boundary-condition scan across both reviewer backends reproduced a previously discovered false success on partial input: four truncation levels passed historically and failed after repair. The scan also found and repaired cancellation during process reaping. In real CLI probes, Claude had no writing tools; Codex attempted writes in five of five read-only trials, each write tool failed, and no disposable repository changed. These tests cover specified paths and versions, not field reliability. A preregistered shadow study of metadata-only review allocation accrued 25 formal observations before an exact-runtime regression found a third defect: a reviewer exiting nonzero with a well-formed verdict was counted as complete. Exit status was not recorded per attempt, so exposure cannot be resolved retrospectively. The 25 formal and two pending records remain an audit cohort; the measurement-valid cohort restarted at zero and collection has begun. No gate result is reported.

We study whether small LLM agents can operate effectively under explicit wall-clock time budgets by both respecting the allocated runtime and using available time productively. We evaluate Qwen3.6-27B on five competitions from MLE-Bench Lite and Qwen3-4B on Zork I (Jericho), two agentic benchmarks where additional computational time can meaningfully improve performance. In the simplest setting, where the budget is stated only in the prompt, agents fail to translate the stated budget into controlled use of time. These failures arise from gaps in time awareness, since the harness provides no timing feedback, but also because they cannot reliably anticipate the duration of actions, and do not have a learned mapping from available time to an appropriate strategy. We investigate two complementary classes of interventions: harness-based mechanisms that expose timing information and enforce deadlines, and reinforcement learning with budget-aware rewards. Injecting timing information through the harness substantially improves budget adherence for Qwen3.6-27B without measurable loss in performance, while enforcement hooks tighten adherence further. RL with GRPO achieves near-perfect budget adherence on Zork I and generalizes to held-out budgets not seen during training, but does not improve task performance over the untrained harness on MLE-Bench. Once agents are made to respect the budget, they still fail to use additional time to improve task performance. RL-trained policies learn when to stop but often fill extra time with repeated actions, and GRPO training on multiple budgets tends to collapse toward the strategy learned for the shortest budget. Our results reveal a gap between time adherence and productive time allocation, which remains a central challenge for budget-conditioned agents.

Self-improving LLM agents can adapt a credit pipeline to a changed rule, but an agent that rewrites itself destroys the artefact a supervisor reviews: a named change, a recorded test, an approval. We argue that self-evolution is reviewable only if it is confined to the runtime harness (instruction text, tool-call logic and primitive composition) while model weights stay fixed, so that every adaptation is a diff with a cause and a test attached. We give a dual-loop engine built on that bound, with one admission gate that writes a hash-chained record before deployment, and we measure the gate in simulation, with a simulated agent and a seeded-search proposer rather than language models. Across three families of supervisory re-interpretation at three severities, 10 seeds each, the gated loop admitted 144 of 7,449 candidate changes, none of which worsened error on held-out history, and restored the false-positive rate to the oracle level without raising missed flags in every low- and mid-severity cell. With the gate replaced by the check an unbounded system applies (fewer errors visible in recent traces), the same loops admitted 309 harmful changes and left missed flags above 10% in 49 of 90 runs: false positives fell because the screen was loosened. Evaluated on pre-shift labels, the gate rejected every candidate, so a re-interpretation must be encoded as a rule that relabels history. Parametric and scope shifts were repaired locally, a structural one only by primitive replacement; at the highest structural severity the gate's fixed tolerance blocked the correct replacement in half the seeds. We map the mechanisms to the EU AI Act's provisions for high-risk credit scoring and note that the April 2026 US model-risk guidance excludes agentic AI from its scope.

Computer-use agents (CUAs) perform tasks across applications (such as desktops, mobile apps, and web browsers) by observing graphical interfaces and issuing commands such as clicks and keystrokes. These interfaces combine trusted controls and content with untrusted content needed for legitimate tasks. An adversary controlling this untrusted content can embed instructions or misleading visual cues to change the agent's intended action or redirect its commands to the wrong interface target. We formalize security requirements for both the agent's decisions and their execution through GUI commands. In an ideal execution model, we show that enforcing both requirements at each step protects execution traces. We instantiate this model in Secure-CUA, our system for secure CUA execution. Its key idea is to commit to an explicit per-action program, called an $\textit{action transaction}$, before accessing untrusted content. Each transaction fixes its queries to untrusted content and the permitted uses of their responses. The system masks untrusted regions and evaluates each transaction to produce the next action, using an isolated query model to answer its queries. It then locates the intended interface target using the masked interface. Under the model's assumptions, Secure-CUA is secure by design, while generating a new transaction at each step helps maintain high task utility by adapting to changing interfaces. We evaluate Secure-CUA under benign conditions on 400 WebArena tasks using three frontier models across $5$ seeds, yielding $6,000$ execution traces. Secure-CUA achieves an average task success rate of $53.55\%$, compared with $55.12\%$ for Vanilla-CUA and $13.17\%$ for CaMeL-CUA.

Supervisory governors can interfere with the tool-using agents they regulate. We study this possibility in a controlled file-recovery environment where increases in regulatory intensity trigger experimentally imposed tool failures. A cost-blind governor can turn these failures into persistent blocking that prevents task completion. We compare this governor with a backoff rule that reduces intervention probability using a moving average of known induced events. On a hand-coded stochastic-policy agent, the failure pattern appears under both result replacement and execution of corrupted tool arguments. For the persistent policy, adaptive backoff improves completion relative to a fixed weak governor with approximately matched intervention frequency. A Gemini 2.5 Flash experiment comprising 576 episodes across 6 tasks also shows reduced blocking and improved completion under backoff; among the tested settings, intermediate backoff strength achieves the highest observed aggregate success. These results identify an interaction between intervention cost and persistent action blocking, together with a possible mitigation. The cost mechanisms are imposed and their induced events are directly observable to the backoff rule; applicability beyond this controlled environment remains an empirical question.

LLM-based agents are entering decision-support roles in defence staff work, where the obligations they must respect are already written down and binding, and where retraining is not available as a control because models arrive as procured components. What can be placed under engineering control is the interface between the agent and the systems it acts on. Those obligations are at once spatial, temporal and text-semantic, and a violation typically lives in the composition of a multi-step interaction, which is why per-event guardrails miss sequential tool-attack chains. We present a multi-aspect runtime-verification framework that decomposes a natural-language policy clause into a typed spatial/temporal/semantic triple over one canonical event stream, checks each aspect with its own monitoring specification, and fuses the verdicts through a four-valued algebra that carries provenance. The spatial aspect is interpreted over a weighted two-sorted location graph in which mission geometry and information-release topology are one object; we show that these spatial obligations are not in general subsumed by a first-order temporal specification. The past-time aspect runs on the unmodified MonPoly engine, which agrees with our reference monitor at every time point. Across two mission domains, casualty evacuation and contested sustainment, and one civil domain, composition under the precautionary blocking policy drives attack success to zero with no observed false positives and microsecond-scale per-event cost, while every single aspect and every pair leaves a substantial share of attacks succeeding. In a closed-loop experiment a policy-naive planner reaches a violating state in most unshielded missions and in none when shielded, and four refused episodes in five still recover to a compliant outcome.

Agentic coding makes code generation cheap, but reliable completion remains difficult: the agent that writes the code is a weak judge of whether it is done. We present Humanize, a multi-agent orchestration workflow for agentic coding built around judgement engineering: explicit, mechanically enforced decisions at the boundaries between planning, implementation, review, and learning. A human approves a plan contract, a builder agent implements it in rounds, and a reviewer agent from another vendor decides completion; deterministic hooks, not a model, route work between these roles and enforce 72 mechanical gates. Viewed as a Markov chain over repository states, alternating builder and reviewer samples jointly from two models, so a defect survives only if both miss it. We study Humanize through its deployment, 118 public postmortems of real loops, and its applications. Over 68 versions in 108 days, it gathered 1,468 GitHub stars. Applications include a 567-file gem5 build-system migration under upstream review; Kernel Design Agents, which extend the loop with a kernel knowledge base and profiling feedback and placed in the top three of all three Full-Agent tracks of the MLSys 2026 FlashInfer contest; and, through Humanize Olympiad Agents (HOA), full scores in IOI 2026, IMO 2026, IPhO 2026, and IBO 2024, 418.5/437 in IChO 2026 (gold-medal). Humanize also achieves 672/672 on PutnamBench and ranks first (251/303) on Lean-Eval's leaderboard even competiting with professional mathematicians. The postmortems show that independent review catches unsupported builder claims, but stopping remains a key weakness. In reports that separate rounds by phase, two thirds of rounds occurred after implementation was accepted. This evidence is observational, not a controlled comparison of workflows.

LLM tool-use agents operate in dynamic environments where many actions carry operational risk. However, most safety mechanisms react only after errors manifest. Existing pre-emptive approaches either fine-tune the agent on chain-of-thought deliberation or compile natural-language guardrails into runtime checks, but they do so without exposing a structural, auditable verdict. We propose POLAR, a guardrail framework for small tool-calling agents that assesses reversibility through a structured two-layer ontology. POLAR assigns each action a graded reversibility score by deriving a candidate inverse sequence; calls failing a threshold are pruned before execution. Evaluated on $τ^2$-bench across six agent models, POLAR improves mean task reward by 0.11 to 0.18 points on airline for four of six agents, but only eight of eighteen model--domain cells improve overall; retail and stronger agents often regress. POLAR provides an auditable structural check and characterizes its task-utility trade-offs. Reward is not a direct measure of prevented harm.

Many agent runtimes give the agent a tool for editing its own configuration. Some of that configuration grants abilities, such as enabling a tool. Other parts set the agent's limits: which directories it may write to, who may send it messages, which network address it listens on, how callers authenticate, and the gate that blocks risky writes. If the agent can edit those limits, a single ordinary request can widen them. We study this in a deployed, model-agnostic runtime. We propose a rule: the agent may change fields that grant abilities, and may never change fields that set its limits. We enforce the rule as a containment floor inside the configuration tool and measure what happens with and without it. Without the floor, a frontier model wrote a protected value on 25 of 72 ordinary requests that gave it permission to change settings, often when the request never named the field. Prohibitions written in the system prompt failed in a predictable way. A prompt that listed the protected field names stopped every request that used those names (0 of 36 saved, against 17 of 36 with no prompt) and did not stop the requests that only described the goal (10 of 36 saved, against 8 of 36). A prompt that described the forbidden effects did the reverse. With the floor, 0 of 167 protected writes were saved, although the models attempted a protected write in 65 of those cases. A search for other routes through the tool found only one, a pinned shell, which the floor's scope statement already excludes. The study covers two models and a single agent. We state what that does and does not support.

Language-model agents are evolving into long-running services that interact with models, tools, computers, mobile devices, and distributed environments. Existing agent frameworks simplify reasoning and tool invocation. However, cloud-centric designs face three limitations: centralized execution increases failure impact, scaling pressure, and compute cost; extending agents across computers, mobile devices, and edge environments requires a unified execution abstraction with permission control; and long-running executions require consistent lifecycle management across failures, recovery, results, usage, and settlement. We present Nexus, a cloud-edge platform that treats each invocation as a persistent task. Nexus uses an OpenWrt-based runtime for distributed serving, run-scoped delegation for authorized access to Computer and Mobile environments, and persistent records to track execution, outputs, failures, recovery, usage, and charging across cloud and edge components. We evaluate Nexus on controlled, cross-device, and model-driven workloads. All ten Computer-Android workflows succeed, and all six revocation tests block subsequent writes while preserving prior authorized reads. Under worker loss, journaling eliminates duplicate appends (six to zero per task), adding 0.933 s mean normal-path overhead. Across 24 matched task pairs, Nexus completes 24 tasks versus Dify's 22 and is a median 3.88 s faster on jointly successful pairs. In a separate workload, Nexus operates under a smaller tested incremental-runtime memory ceiling than Dapr (16 versus 64 MiB), although Dapr achieves lower successful-call latency. These results demonstrate how locality, operation-scoped authority, and persistent result identity support cloud-edge agent services with workload-dependent costs.

Indirect prompt injection (IPI) hides adversarial instructions in content that large language model (LLM) agents read at runtime. As agents compose heterogeneous capability units, including Tools, MCP servers, and Skills, the carriers of injection multiply, and defenses built to recognize attack patterns fall behind them. We instead shift defense from covering attack patterns to one stable point: whatever the carrier and however the injection propagates, harm materializes only at the \emph{execution boundary}, where the agent turns internal state into an external action or released output. Safety there turns on two conditions, both settled by the trusted task rather than by the run: whether the proposed effect is authorized, and whether the runtime information reaching it is endorsed by that task. We present APEX, an active defense that enforces both at this boundary from a single authorization contract compiled before untrusted execution: \emph{evidence-gated prevention} admits an effect only when the contract justifies it, while \emph{deception-based exposure} makes unendorsed use reveal itself before the effect commits. Protection therefore follows from what the task permits rather than from how an attack is built, and applies uniformly across capability units without attack-specific policies or taint tracking. Against 13 baselines, APEX attains 0% attack success on five of six benchmarks and 0.56% on the sixth, holds 0% under adaptive attacks on all three capability-unit types, and remains effective across defender backbones. Code is available at https://github.com/ZhengXR930/APEX_official/tree/official.

Tool-using large language model (LLM) agents turn generated text into real side effects, so poisoned tool metadata, retrieved pages, memory, and reusable skills can steer the next call. Vetting an artifact before admission does not settle this. A safe variant and a leaking variant can produce the same admission evidence, and a sound gate then cannot relax that site for either. We make that condition precise, which leaves the last boundary a deployment can still act on. We present Provenance-Aware Capability Enforcement (PACE), which mediates every tool call immediately before it executes. Path confinement proposes an executable cut of represented influence paths, while capability and effect verification checks schema-defined effects against authority compiled from the authenticated request. We distinguish the certified execution contract from the evaluated configuration, which can restore an authorized call after a proposed block or apply a declared repair. Confinement requires the final action to preserve the certified cut. On eight executable agent-security benchmarks with three target-model families, the evaluated configuration gives strictly lowest attack success in 62 of 79 eligible attack columns and ties in 14; full-benchmark native utility loses at most three points relative to the undefended agent. A complete ablation over 1167 paired cases attributes most security gains to effect verification and refusal control to boundary adaptation. A reduced-scale adaptive search succeeds on 0/30 out-of-authority targets against the defense.

Contextual security defenses prevent AI agents from taking rogue actions by synthesizing a task-specific policy and enforcing it on the agent's tool calls. In multi-step tasks, however, which actions are valid often depends on what the agent has already done and learned. We present Sapien, a policy engine for enforcing stateful contextual policies. A Sapien policy specifies permitted tool-call sequences using a regular expression extended with stateful predicates, deferred policy generation, and scoped semantic checks. We show that Sapien stays within a few percent of an unconstrained agent's utility. Even if the agent is fully hijacked, Sapien's policies rule out 93-95% of attacks on AgentDojo and 62-85% on Toolathlon (twice as many as tool allowlists on long-horizon tasks).

Causal action verifiers gate an agent's state-changing tool calls by checking whether each proposed intervention is identifiable against a committed action-state graph, and they issue a certificate that carries the identification argument and a one-sided lower confidence bound. One such verifier, CIVeX, reports zero false executions on a confounded tool-use benchmark. We red-team it by corrupting only the committed graph. Omitting a single bidirected edge takes it from zero false executions to 15.3% at the benchmark's published confounding strength, with 91% of its executions harmful and utility falling from +2.27 to +0.35. Reversing one arrowhead, so that a mediator is committed as a confounder, gives 48.9% false executions and no correct ones. Every one of these actions carries an internally valid certificate. An attestation step that tests each observationally certified execution against a bounded randomised sample detected both attacks, with 2 false alarms in 555 executions on a truthful graph; refusing what fails the test, or cannot be tested, gave zero false executions in every setting we measured. It does not restore beneficial execution: at the published strength 97.1% of beneficial actions are still never executed, because the same misspecification rejects them before attestation runs. Those rejections carry certificates too, and auditing them works, but its cost scales with the number of rejections rather than the number of executions. Recovering safety costs 127 experiments per 1,050 actions; recovering the lost value costs 614 more, at which point the audited verifier makes the honest graph's decisions on every instance and spends exactly its experiment budget. An audit that inspects only executions protects against wrongful action. Wrongful inaction has to be paid for separately.

Coding agents solve repository-level tasks through sequences of actions, where a single erroneous action can misdirect subsequent decisions and increase recovery costs. Existing approaches use execution feedback for recovery or specialized checks to block errors, but deciding before execution whether intervention will benefit eventual task completion remains challenging. To address this challenge, we propose HiSentinel, a hindsight-distillation framework that trains lightweight 0.6B and 1.7B sentinels to select pre-execution interventions aimed at improving task completion rather than correcting every imperfect action. A privileged teacher uses recorded execution outcomes as evidence for intervention judgments, which are distilled into a causal student that receives only the pre-action context and proposed action. Beyond identifying whether and when to intervene, the sentinel must also provide actionable feedback that helps the coding agent recover or obtain necessary human input. To support these capabilities, we introduce SWE-Intervene, an action-level dataset constructed from software-engineering trajectories that annotates whether an action should be allowed, autonomously redirected, or paused for human assistance, together with corresponding intervention feedback. Across SWE-bench Verified Mini and Ask or Assume, HiSentinel consistently improves task completion across Sentinel scales and coding-agent families, with gains of up to 14% and 10%, respectively, while maintaining competitive token consumption. These results demonstrate that lightweight pre-execution intervention can effectively prevent error propagation and improve the reliability of autonomous coding agents.

Benchmarks, audits, and agent protocols describe performance, permissions, and repair, but not how observed evidence should change an agent's authority during a consequential task. We call this the assurance-transition gap. We propose a Runtime Assurance Contract (RAC), a policy-level formal schema binding autonomy boundaries, component eligibility, evidence state, transition policy, human-review capacity, and non-compensatory gates. Under RAC, soft metrics may inform routing, whereas a failed or unknown mandatory gate forces retry, switch, escalation, deferral, or stop; aggregate performance cannot authorize action. We define the contract, an evidence record, a permission rule, and five invariants, and illustrate them in clinical, industrial, and judicial failure probes. We then report a deterministic failure-injection study in agentic coding: 280 constructed cases evaluated by a gate conjunction, a score-only rule, and a restricted protocol baseline. At the published example weights and threshold, the score rule admits 80 of 100 block-required injections and all 40 review-required injections. Tuned in hindsight, it matches the conjunction on this corpus. For positive weights, a positive threshold, binary risk signals, zero-signal controls, and an injected case firing each signal alone, we show that exact agreement holds if and only if the threshold does not exceed the smallest weight. A separate set of 18 hand-authored traces checks version-pinned evidence and review transitions against simpler policy variants. In a further prospective synthetic holdout of 24 episodes, two blinded LLM judges assign identical labels to all 72 action attempts; RAC and a separately implemented full stateful baseline both match these labels. These studies test mechanisms on synthetic cases; they establish neither deployed safety nor cross-domain effectiveness.

LLM-based agents extend their capabilities through third-party skills that provide task-specific instructions, scripts, and tool-use procedures. However, malicious instructions inserted into an otherwise benign skill can cause a benign user request to trigger dangerous Tool Calls, including data exfiltration, file deletion, or unauthorized code execution. This paper presents ActionGuard, which inspects skill-influenced Tool Calls immediately before execution. ActionGuard separates the target agent's action-generation context from the safeguard's authorization context. The target agent may use the original skill for planning, but the Reviewer does not receive the potentially poisoned raw skill text. Instead, it determines whether each action is justified by the trusted user request using a balanced skill profile, current and recent Tool Calls, and local script contents. ActionGuard intercepts each Tool Call at OpenClaw's before-tool-call stage and enforces the Reviewer's ALLOW or DENY decision under a fail-closed policy. We evaluate ActionGuard on 139 contextual and 180 obvious injections in a SKILL-INJECT-based setting against Dynamic Guardian and SkillGuard, using three open-source and two commercial Reviewer models. Each condition is repeated three times and evaluated using Attack Success Rate (ASR) and Task Success Rate (TSR). Overall, ActionGuard reduced ASR by 35.54 to 46.11 percent relative to existing safeguards and by 70.44 percent relative to No Safeguard, while maintaining high benign-task completion. These results show that execution-boundary authorization grounded in trusted user intent and runtime evidence can restrict unauthorized Tool Calls induced by skill injection.

Before a validator can be promoted to a hard gate on a deployment pipeline, it has to be shown that its firing separates outputs that reach users in working order from those that do not. We run that screen on 13 validators in a deployed generative agent, against 550 runtime and 350 static builds labelled by downstream outcome, and report each check's marginal separation $J=\mathrm{TPR}-\mathrm{FPR}$ with Newcombe intervals and Fisher exact tests. Two checks survive correction for multiple comparisons, two more are nominal only, and the remaining nine are not distinguishable from zero, three of them because they never fired on any sampled build. Execution itself is not random with respect to the property being gated, and this replicates: across four runs covering 1,867 builds and ten distinct runtime checks, probes were skipped on 144 of 895 broken builds and 1 of 972 acceptable builds (per-run rates 15.6% to 16.6% against at most 0.3%), every skip carrying the same unsafe-to-probe reason. Because a skipped check is recorded as a pass, this imposes a ceiling that no check quality can lift: a check that needs a live artifact cannot operationally detect more than about 84% of broken builds in this harness. For the one check with construct-specific labels, a detector built for blank output fires on 0 of 90 human-labelled blank builds (95% upper bound on sensitivity 3.3%), and the global frame statistic it approximates separates the classes only weakly (AUC 0.59), so the gap is not a threshold that needs tuning. The same gap appears one layer up: on a census of tens of thousands of judge-scored builds, 32.5% of rejections carry no recorded issue at all. We argue that evaluation records must distinguish a check that ran and passed from one that did not run, must carry the evidence for a rejection, and that an inventory of checks is not evidence about a gate.

Tool-using LLM agents remain vulnerable to indirect prompt injection because trusted instructions and untrusted observations share one context, allowing malicious content to steer consequential input-filtering defenses. Multi-path consensus defenses still leave a high attack success rate because they examine content or aggregated outputs rather than authorizing effects, especially for the within-tool attack, which preserves the intended tool but manipulates its arguments. Data-Flow Control such as CaMeL provides stronger guarantees, but incurs substantial time latency that limits practical deployment. We introduce ToolFence, which compiles a typed authorization blueprint before execution, enforces it through a deterministic monitor, and when the blueprint is incomplete asks a judge to grant new capabilities rather than adjudicate each concrete call. ToolFence provides two key advantages. First, its fine-grained provenance-aware authorization enables the system to distinguish user-authorized values from untrusted observations, effectively addressing the within-tool attack. Second, its deterministic fast path and capability-level runtime grants substantially reduce the frequency of expensive judge calls, improving runtime efficiency. On AgentDojo with Qwen3-max, ToolFence reduces overall ASR to near zero with only a 3.80 percentage-point clean-utility drop and practical runtime overhead.

Many computer tasks recur: the same workflow runs many times, with new inputs and from different starting states. Current computer-use agents re-plan every step of every run, which makes them costly and unreliable on such tasks. We introduce neuro-symbolic computer use, in which a recurring workflow is executed by a learned policy rather than re-derived by an agent on each run. The policy fixes the decisions that are stable across runs (ordering, variables, loops, and branches) in executable code, and delegates observation-dependent decisions, such as grounding and state checks, to neural models. We learn these policies with neuro-symbolic policy iteration: starting from one agent trajectory, it executes the policy, diagnoses failures with task-completion and step-level judges, and revises the code with a coding model informed by an agent's continuation from the point of failure, without access to the benchmark evaluator. Iterating on generated parameter and initial-state variants makes the policy reusable, and a pre-action verifier guards each state-mutating step at deployment. On OSWorld-Verified and ScienceBoard, the learned policies achieve the highest Pass^3 of all methods in all four settings, 3.6-15.8 points above the base agent, while cutting per-run cost by 15-217$\times$ and latency by 3.4-5.1$\times$. On OSWorld-Verified, policies built only on variants transfer to the held-out original tasks, exceeding AutoRPA by 8.6-17.5 points in Pass^3.

The compiler must read modules a physics-based solver cannot build without; the coding agent must not read that intellectual property. The harness does not ship that rule. We classified fifteen read routes against a container, permission rules and a sandbox. None of the three can tell which program is reading.

We describe AfS (Agent for Science), a platform built for long-horizon scientific work, where a project runs for tens of hours across dozens of agent runs with a human present only occasionally. Most agents for science are general coding agents with a skills folder attached, and they inherit that lineage's failure mode: under pressure to finish, they fabricate, skip, or smooth over. Our design rests on one claim: most of the credibility of machine-made research can be moved from asking the model to behave to making the non-compliant state unrepresentable. We encode research discipline as mechanically enforced laws (commitment before measurement; unforgeable freezing; reports are not facts; evidence persists but verdicts do not; negative results are first-class; mechanical questions to the framework and semantic judgment to the model), organized around three time horizons: a minimal set of research nodes within a run, an inquiry contract with frozen closure conditions and a hash-chained artifact ledger within a project, and a two-tier knowledge base with promotion by rewriting across projects. This is a system description written under one rule: each mechanism appears in exactly one place, with the invariant it enforces, the failure it prevents, the way it is realized, and the cost it imposes. It covers the node contract, the write-path gates, the two-tier memory, and the runtime substrate. Three traces walk real failure attempts through the mechanisms that catch them, and two closed campaigns are included as worked illustrations rather than as an evaluation. We report no benchmark: a process-integrity suite that would support quantitative comparison is under construction, and what we can measure today is only the operating cost of the machinery.

Tool-enabled agents form calls from model-visible interfaces, while hosts later select their implementation. Standard dispatch omits the descriptor-handler relation. An unchanged and schema-valid call can therefore acquire a different security effect during rollout, reconnect, or delayed approval. We call this failure schema-epoch drift. We present formation-consistent dispatch (FCD), which connects implementation analysis to execution authority. Reviewed profiles produce provenance-bound over-approximations of declared in-scope effects from official source. Under a closed-target approval policy, a verifier applies each formed call to a summary and captures a successor only when its effects fit the call's security contract. Atomic admission and a final-hop fence preserve this decision to the effect. The exact source retains priority, and the captured successor becomes eligible only after source retirement. Stock releases and deployment changes reproduced the failure. Four profiles covered 32 official releases: 29 required no release-specific change and three escalated. A frozen 16-release expansion matched a separate source oracle. In a preregistered stock comparison, FCD completed all three pending calls whose effect remained private and blocked all three whose omission became public. Exact pinning and release-wide denial stopped all six calls, while release-wide approval completed all six but produced three public effects. A separate lifecycle experiment carried a formation-captured certificate across source retirement. The same safe certificate installed later governed new formations without expanding the pending call's authority.

LLM-based Web agents can autonomously complete user tasks, yet deceptive interfaces can steer them toward outcomes that conflict with users' interests. Existing defenses primarily intervene on agent behavior through blocking, guidance, or replanning. We identify a distinct failure mode: a task-valid action can still realize an unauthorized consequence because of the current Web state. This motivates treating task-relevant Web state itself as a runtime control target. We introduce Veer, an agent-side runtime defense that leaves task planning to the base agent and intervenes on Web state when a proposed action would produce an unauthorized consequence. Before modifying the live environment, Veer constructs a prospective intervention trajectory toward a safe task-relevant state and executes it with runtime grounding and verification. Across TrickyArena and WebDecept, Veer achieves the highest safe task completion in all three evaluation settings, exceeding the next-best defense by 15.9 and 25.0 percentage points on TrickyArena-Single and TrickyArena-Multi, respectively, while reducing dark-pattern success on WebDecept to 0.3%. These gains persist across dark-pattern types and all 12 agent, model, and benchmark configurations. Ablations show that active state intervention provides the largest gain, while prospective rollout and temporal evidence contribute additional improvements. These results establish task-relevant Web state as an effective runtime control target for protecting Web agents from deceptive outcomes.

We propose RegLLM, a diagnostic harness for bounded autonomy in regulated agentic workflows. It instruments six trustworthiness signals: citation validity, source grounding, schema compliance, escalation correctness, constitutional alignment, and unsafe-action rate. Signals are distinguished by their source of supervision: programmatic verifiers, task-level escalation labels, or AI-judge scores. A deterministic runtime supervisor blocks ungrounded answers and forces escalation, logging interventions. The same domain constitution informs evaluation, training rewards, and serving guardrails. Task-level should-escalate labels make the act-versus-defer decision a measurable training signal. We demonstrate the harness at smoke scale. An offline reference run (n=12) lifts escalation recall from 0 to 0.67 and reduces unsafe-action rate from 0.33 to 0.08 when governance is enabled. Two single-GPU Qwen2.5-3B LoRA/DPO pilots (n=8, same seed and evaluation split) expose substantial variation: nominally identical RL-base configurations yield task success of 0.25 versus 0.12 and escalation recall of 1.0 versus 0.5. An answer-quality adapter changes recall from 1.0 to 0.5 in Run A, but from 0.5 to 1.0 in Run B. An escalation-aware variant produces no measurable change in Run B. These small pilots do not establish reliable adapter effects or production readiness. Their contribution is diagnostic: configuration variance can overwhelm apparent tuning effects on bounded-autonomy metrics, motivating larger evaluation sets and repeated runs.

Language-model agents increasingly use tools to act on external systems. Earlier actions can alter files, permissions, database records, or other state, making a later routine-looking action harmful. Yet the visible interaction may not reveal the underlying state needed to assess that action. We formulate attack and defense as partially observed state control in SEAD, deriving their design requirements from this shared execution process. Because attackers supply instructions while the target chooses concrete actions, DART decomposes harmful goals into locally plausible steps and uses feedback from actual tool execution to guide trajectory search. The defender must decide before execution with incomplete state evidence. SAGE can therefore investigate relevant state through read-only queries before allowing or blocking each action, including those proposed after a block. We construct an environment-verifiable dataset integrating controlled initial states, replayable tool environments, and task-specific executable checks. Across four target models, DART improves semantic attack success by 18.8--35.9 percentage points over the competing baseline, with consistent gains under executable verification. On recorded trajectories, SAGE preserves 95.79% of benign trajectories while intercepting 92.73% of harmful paths by the harm-enabling boundary. In online attack-defense evaluation, it reduces DART's executable attack success from 48.0% to 4.0%. SAGE remains effective across four attack methods and generalizes to out-of-domain environments. Our code and data is available at https://github.com/EverywhereSafety/SEAD.