LLM Red Teaming

Autonomous red team agents increasingly stress-test AI-enabled cyber defenses by planning strategy and executing multistage attacks. Reinforcement learning (RL) and large language models (LLMs) offer complementary mechanisms for the planning and execution such agents require, and prior work has combined them in hybrid hierarchies. Yet a given architecture is typically developed and evaluated within a single environment, leaving open whether an observed advantage reflects a generally stronger decision mechanism or merely alignment with a particular setting. We address this gap with a controlled cross-environment comparison of two homogeneous hierarchical red team architectures: an RL planner with an RL executor (RL+RL) and an LLM planner with an LLM executor (LLM+LLM). We evaluate both against expert autonomous defenders in CybORG CAGE-4 and in Cyberwheel at two network scales, across 18 configurations under one unified disruption metric. We find a pronounced environment-dependent inversion. RL+RL wins the compact, densely rewarded CAGE-4 (78.5% disruption success versus 18.0% for the strongest LLM configuration) and the 100-host Cyberwheel network (81.0% versus 50.5%), while a pretrained cybersecurity LLM agent wins the larger, escalation-gated 1010-host Cyberwheel network (55.0% versus 0.0% for RL). A kill-chain analysis explains the inversion through architecture-specific bottlenecks that aggregate success rates conceal.In the 1010-host Cyberwheel network, RL discovers and compromises hosts but stalls at privilege escalation, whereas in CAGE-4, LLM agents obtain privileged access but rarely convert it into operational impact. These results indicate that conclusions drawn in a single environment may not generalize, and that hybrid planner-executor designs should be motivated by specific failure modes rather than the assumption that one architecture is universally preferable.

Prompt injection is a leading security risk for LLMs and LLM-based applications such as agents. State-of-the-art red-teaming methods for prompt injection leverage reinforcement learning (RL) to train an attacker LLM to generate effective injected prompts. However, when targeting frontier LLMs such as GPT-6-Luna, a major challenge is the cold-start problem: every attack attempt by the attacker LLM fails and thus receives zero reward, providing no signal for learning. In this work, we propose a curriculum learning-based method to address the cold-start problem. In particular, we propose to train the attacker LLM against a sequence of increasingly robust target LLMs, with each stage warm-starting from the attacker LLM obtained in the previous one. However, simply training against a weak target (e.g., GPT-4o-mini) may not sufficiently prepare the attacker LLM to obtain useful learning signals against a frontier LLM (e.g., GPT-5.6-Terra). Instead, we find that the design of the curriculum is critical: after each stage, the attacker LLM needs to partially succeed against the next target LLM such that it can learn from successful attempts to attack the new target. Our extensive evaluation shows that our method can effectively red-team frontier LLMs, achieving an attack success rate (ASR@10) of 93.8% and 45.0% against GPT-5.6-Luna and GPT-5.6-Terra on AgentDyn, whereas state-of-the-art RL methods such as RL-Hammer and PISmith achieve 0% ASR under the same setting. Moreover, we find that the attacker LLM transfers across targets, e.g., an attacker LLM trained to defeat one strong LLM (GPT-5.6-Terra) also succeeds against six other frontier LLMs (e.g., GPT-6-Luna) it was never trained on. Our code is available at https://github.com/albert-y1n/PIForge.

Automated red-team attacks and blue-team defenses for large language models (LLMs) are advancing quickly. However, attackers and defenders are built and tested in isolation, and the resulting scores are hard to trust. To tackle this, we present ACEA (Adversarial Co-Evolution Arena), a platform that connects a pluggable red-team adapter and a pluggable blue-team adapter to a shared target LLM and scores their attack and defense rates with an LLM judge. ACEA contributes four components. First, a pluggable, model-agnostic arena. Any red or blue project connects over a minimal HTTP protocol, which we call the ACEA Standard Adapter Protocol (ASAP). It can be written in any language, and a project that exposes nothing but the protocol is a full participant. Second, an evaluation methodology built for adversarial rounds. Seeding the target with canonical secrets gives verifiable ground truth that separates real leakage from hallucination. We also send each attack to the target even when the defense blocks it, which measures the attack's raw potency independently of whether it was stopped. Together these yield a per-round decomposition of attack strength and defense effectiveness. Third, a real-time, game-style visualization with a detailed end-of-battle report that localizes each failure. The evaluation thus becomes an actionable signal for improving a red or blue project. Fourth, an optional in-context improvement loop that turns each round's outcome into advisory hints for the next. An adapter can then adapt across rounds without keeping state, provided it reads the hints. We describe the design of ACEA and the metrics through which red and blue teams are scored head to head.

Frontier language models that refuse harmful single-turn prompts often comply when the same intent is reached gradually over many turns, making multi-turn attacks one of the least understood failure modes of large language models. Most automated red-teaming methods treat this as a generation problem: produce attacks that break the model. We argue it is better framed as a search problem: discover, organize, and iteratively refine a diverse archive of attack strategies, producing a structured map of how a target model fails rather than a list of one-off successes. We introduce EvoFlint, which applies evolutionary quality-diversity search to multi-turn red-teaming. Attack strategies are phased conversation plans, not raw prompts, and are evolved through LLM-driven mutation and crossover. A Pareto fitness over attack success rate and peak severity preserves selection signal from near-miss attacks. A risk-indexed archive runs novelty search with local competition over strategy description embeddings inside each cell, maintaining diversity without committing to a predefined style taxonomy. A generation-level memory accumulates target-model insights across the population and feeds them back into strategy generation. On the HarmBench-test split, EvoFlint reaches attack success rates of 35.8% on Claude Sonnet 4.6, 59.7% on GPT-5.4, and 94.3% on Qwen3-32B, alongside 98.7% on the older GPT-4o included as a baseline reference. The resulting archive, organized by risk category, exposes for each target which categories of harm its safety training has and has not covered.

Large language model (LLM) agents combine language-based reasoning with external tools to perform complex tasks. Adversarial inputs can exploit interactions between the agent and its environment, causing the agent to violate safety policies during execution. Yet existing evaluations often reduce agent safety to a single attack success rate (ASR), collapsing exposure, execution, observation, and adjudication and potentially conflating actual violations with evidence visibility. We introduce REDAgentBench, an executable framework for autonomous red-teaming and faithful measurement. It derives attacks from explicit safety constraints and associated agent-system vulnerabilities, runs them in isolated service sandboxes, and verifies harmful effects from service receipts and final-state changes. The benchmark contains 1,661 cases across five service surfaces. Across six models and three agent harnesses, macro-average ASR is 65.69%; reported ASR varies with harness and evidence view, while evaluation-context disclosure changes execution behavior. In a state-grounded diagnostic cohort, almost one in five confirmed violations with resolved action anchors occurs after the agent states the relevant constraint or risk, revealing a Recognition--Execution Gap. Finally, a training-free policy reminder reduces confirmed violations by more than 70 percentage points in matched replay. These findings show that executable evaluation can improve safety measurement and identify actionable intervention points.

Autonomous cyber defense systems based on Deep Reinforcement Learning (DRL) have attracted significant research attention, yet remain evaluated almost exclusively against static, heuristic red agents, leaving their robustness against adaptive threats critically understudied. Meanwhile, recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have improved LLM reasoning, but their integration into cybersecurity remains elusive due to the absence of suitable benchmark environments and interaction datasets. To bridge this gap, we introduce Trident, an agentic LLM red teaming framework comprising three components: a dynamic benchmark with isolated sandbox servers spanning CybORG CAGE 4 and CyberWheel, a dataset comprises over 13,000 high-fidelity red-blue interaction trajectories for RLVR, and a ``Code-as-Policy'' RLVR agentic architecture Trident Agentic). The latter reformulates red agent training as a contextual bandit via a tripartite Log Summarizer--Planner--Coder design, where a trainable Planner generates complete attack strategies from compressed execution logs, which a frozen Coder translates into executable Python policies deployed against live DRL defenders. Empirical evaluations reveal a fundamental brittleness in existing defenses: with a single trainable 7B planner, Trident reduces blue agent defensive performance by an average of 522% compared to static red agent baselines while autonomously discovering emergent behaviors such as decoy avoidance and adaptive state prioritization that static heuristics entirely fail to uncover.

We introduce \textbf{GPT-Red}, an automated red-teaming agent that is trained to discover novel prompt injection attacks against frontier LLMs. The goal of this model is to evaluate and improve the robustness of our production systems. To this end, we use it to adversarially train GPT-5.6, our most robust model to prompt injections to date. To create GPT-Red, we design a scalable self-play algorithm where the model is tasked with attacking a diverse population of simultaneously-trained defender agents. We train the model on realistic red-teaming environments using compute on the same scale as some of our largest RL post-training runs, making it the single-largest LLM safety training run ever documented. GPT-Red excels at red-teaming: it reliably breaks our past models up to GPT-5.5, it finds more successful attacks than human red-teamers, and it generalizes to held-out environments, defender models, and harnesses. In the future, we expect that as we improve the robustness of each new GPT model, it will in turn will provide better learning signal for \textit{even stronger} red-teamer agents, thus unlocking a self-improvement flywheel.

Red-team evaluations of AI models support some claims and not others, and the boundary between the two is calculable rather than merely a matter of judgment. We define the evidential ceiling of an evaluation as the largest factor by which one result can move belief under a fixed testing budget, derive it in closed form for the benchmark null result, and use it to locate that boundary exactly. We find that above a calculable harm rate, a benchmark of modest size certifies a category to a stated evidentiary standard, and a clean sheet is then the stronger of the two possible observations, outweighing a single reproduced failure. Below that rate, no passive benchmark of feasible size provides the specified evidence of safety under the fixed scoring rule and approximately independent trial structure. The crossing between the two regimes has a closed form. The bound is not specific to benchmarks: written in terms of a procedure's hypothesis conditioned elicitation rates, it covers adaptive and automated red teaming as well, and shows that discrimination between the hypotheses rather than attack success is what determines evidential worth. Auditing eight evaluation suites against the boundary, we find that current benchmarks are adequate for high-frequency harm categories and several orders of magnitude short for rare, catastrophic ones. Safety benchmarks are not uninformative. They are informative about a specific and computable set of propositions, and the discipline they need is to state which.

Safety evaluation of large language models (LLMs) relies largely on single-turn attack datasets and single-judge scoring, underestimating risk from adaptive multi-turn adversaries and reporting a single success rate that does not separate partially actionable outputs from those carrying complete operational detail. We propose AMT-X (Adaptive Multi-Turn Exploitation), a phase-structured multi-turn red-teaming framework. Unlike prior multi-turn attacks that rely on ad hoc escalation or free-form per-goal plans, AMT-X casts the attack as an explicit, reproducible multi-phase state machine driven by semantic signals from the victim, and replaces single-judge scoring with a multi-role jury whose phase-conditioned checklists gate success on actionable harm. Across six frontier victim models (queried under their default safety alignment, without added moderation layers) and seven Moderation sub-categories, AMT-X attains overall attack success rates of 97.6-100% under a lenient score threshold, but 66.7-78.6% under a stricter gate requiring complete, real, and operational detail: a gap of up to 33 percentage points between partially and fully actionable harm.

Agentic red-teaming benchmarks report whether an injected agent was compromised as a single bit: the attack succeeded, or it did not. We argue that this binary attack-success rate discards the information a defender most needs, namely how harmful the resulting action was. We introduce an action-graded harm rubric that scores an agent's tool-call trajectory on a seven-level ordinal scale (L0 to L6) according to whether the executed action was reversible, whether it crossed scope to reach another party, and whether it expanded privilege. We compute the scale two ways: a deterministic oracle that reads the trajectory and the attacker's stated goal, and a panel of three frontier language-model judges that read a tag-free account of the same trajectory. Across four victim models and two defenses on the AgentDojo workspace suite, severity grading exposes three cases the binary metric hides, including a defense that reports a zero attack-success rate while still permitting an externally visible cross-scope leak through an unfiltered tool. The judge panel reproduces the oracle with high ordinal agreement (Krippendorff's alpha = 0.91) but shares systematic blind spots that we characterize, most notably a failure to recognize escalation chains. Unlike prior work that provides harm taxonomies, harmful-task completion tests, execution-level safety benchmarks, or severity-aware simulation, our contribution is a reusable, trace-grounded severity instrument applied to the actual actions recorded in existing red-team logs. All code, prompts, and per-episode logs are released.

Recent tools such as OpenClaw have extended the capabilities of LLM-based agents from simple dialog-based systems to fully autonomous agents. These systems allow personalization of the agent through modifiable internal files and the installation of skills. While this enables deployment in a wide range of settings and the automation of diverse tasks, greater capability and autonomy increases the risk of malicious actions being executed unintentionally. In this work, we explore the interplay between agent configuration and the risk of executing dangerous actions without explicit instruction. To this end, we propose CONfiguration Tree-search for Red-teaming Agents (CONTRA), an LLM-assisted tree-search algorithm that discovers agent configurations resulting in the execution of malicious actions. CONTRA works by reasoning about benign yet dangerous configurations and evaluating them in a simulated environment. We construct a dataset of the 473 most popular skills from a public repository, along with 2-5 corresponding malicious target actions per skill. In a large-scale analysis, we find that 75.1% of skills have at least one configuration resulting in the execution of a malicious action, most of which have not been detected as containing malicious content by existing scans. Overall, CONTRA successfully identifies a configuration leading to the execution of the target action in 39.2% of all tested cases. Our findings demonstrate that current agents provide insufficient safety with respect to personalization.

Multimodal agentic retrieval-augmented generation (RAG) systems expand the attack surface beyond prompt injection to include text poisoning, image injection, direct-query attacks, and orchestrator-level tool manipulation. Existing red-teaming approaches are typically surface-specific and often recycle known attack templates; on text-poisoning benchmarks we measure 73-84% exact duplication. We present MIRROR, a unified cross-surface framework that performs memory-guided Monte Carlo tree search while conditioning candidate generation on retrieved context under an explicit novelty constraint. A deterministic Novelty Gate rejects any candidate matching the retrieval set under normalized comparison, allowing retrieval to inform search priors without enabling prompt copying. Across four attack surfaces on a multimodal agentic RAG target, MIRROR attains 76% ASR on image poisoning compared with 52% for baselines, 97% ASR on orchestrator attacks at half the query cost, and the lowest cross-surface variance (coefficient of variation 0.47). In contrast, specialized baselines collapse across surfaces: suffix optimization reaches 79% ASR on text poisoning but 1% on direct queries. We release ART-SafeBench with 41,815 in-package records and runtime adapters yielding 41,991+ total records across four surfaces.

Scaling adversarial evaluation of large language models requires both a method for generating hard inputs and a reliable way to confirm that resulting failures are real. We present AdversaBench, an end-to-end red-teaming pipeline that mutates seed prompts with five structured operators, queries a target model, and confirms failures through a three-judge panel with a meta-judge tiebreaker. We report experiments on 45 seeds across three categories: reasoning, instruction-following, and tool use. Every seed produced a confirmed failure. Four findings stand out. First, operator effectiveness varies sharply by category: inject_distractor scores 0.00 mean reward on instruction-following seeds but 0.80-0.83 on reasoning and tool-use. Second, binary failure rate hides difficulty: instruction-following seeds required 2.4 attacker iterations on average versus 1.1 for other categories, a gap visible in survival curves. Third, pairwise judge agreement of 80-87% coexists with near-zero Cohen's kappa due to label skew; category-level disagreement rates are more informative. Fourth, adversarial prompts generated against Llama 3.1 8B transfer zero-shot to Llama 3.3 70B, suggesting the mutations exploit general behavioral patterns rather than model-specific weaknesses. Code, dataset, and analysis scripts are available at https://github.com/khanak0509/AdversaBench .

Large Language Models remain vulnerable to adversarial prompts that elicit harmful responses, and scaling red-teaming to cover a broad range of failure modes is constrained by the cost of text generation and evaluation. We present \emph{ToxSearch-S}, a speciated extension of toxicity-focused evolutionary prompt search with incremental, embedding-driven niche maintenance, together with an MPI master-worker realization that centralizes population and species bookkeeping on rank~0 while offloading prompt evolution and evaluation to $n_w$ parallel workers. Under a common budget, ToxSearch-S attains peak toxicity competitive with both ToxSearch and RainbowPlus while following a measurably less toxic best-so-far trajectory, indicating lower cumulative search pressure. Diversity is non-uni-dimensional: RainbowPlus yields greater embedding-level spread, whereas ToxSearch-S partitions high-toxicity prompts into more localized behavioral pockets, reflected by a higher DBSCAN cluster count. MPI distribution delivers substantial wall-clock gains, approximately $1.8\times$ with two workers and $3.2\times$ with four, while leaving Best@B statistically indistinguishable from sequential execution. Four-worker runs also produce significantly larger final species cardinality and more toxicity-bearing species, without a reliable gain in global peak toxicity. These results position incremental speciation as a practical quality-diversity mechanism for AI Safety and MPI as an effective means of compressing time-to-result while preserving measured search outcomes.

Production LLMs increasingly rely on toxicity-based moderation filters as a primary defense, assuming that harmful intent correlates with toxic surface wording. We show this assumption is fundamentally brittle: surface toxicity and adversarial intent can be decoupled by replacing as few as five tokens. We present OTTER (Obfuscated Toxicity-Evading Token Evolution for Rewriting), a black-box red-teaming framework requiring only standard API access, directly targeting the practical constraints of industry security audits. Evaluated on 457 AdvBench prompts across four GPT models, OTTER raises average ASR from 7.0% to 84.0%. We further provide the first quantitative analysis of the toxicity--bypass relationship and a per-category breakdown, translating our findings into actionable recommendations for classifier hardening in production deployments.

Large language model (LLM) agents are increasingly proposed as supervisory components for safety-critical systems, yet their robustness under sustained, adaptive adversarial pressure remains poorly characterized. We present NRT-Bench, a benchmark for multi-turn red-teaming of LLM agents acting as operators of a safety-critical system, instantiated in a simulated nuclear power plant control room. A five-role operator team, each backed by a configurable LLM, runs a plant governed by six critical safety functions (CSFs), while adversaries inject messages over four channels in bounded multi-turn sessions with per-turn feedback. Harm is an objective signal rather than LLM-judged text: a run terminates the moment any CSF is lost, attributed to the causing message. Evaluating four frontier operator models under a fixed-attack paired-replay protocol, we find that adaptive multi-turn attacks reliably push the operator team past a safety limit: across the four models, between 8.7% and 12.1% of attack sessions end with the plant losing a critical safety function. Although the four models look almost equally robust by this aggregate rate, their failures barely overlap: of $149$ sessions, none defeat all four models while a third defeat at least one, so vulnerabilities are nearly disjoint across models rather than nested. The effect of added defences is strongly model-dependent: the same guardrail stack or safety-advisor agent that lowers attack success for one model can raise it for another. We release the simulation venue, attack dataset, and replay tooling for reproducible safety evaluation of LLM agents.

Existing safety benchmarks target general adversarial scenarios but miss finance-specific risks. Financial LLMs face regulatory compliance violations, fraud facilitation, and systemic trust erosion that require targeted evaluation. We introduce FinRED, an expert-guided red-teaming framework for financial LLM safety evaluation developed with financial experts. FinRED uses a novel two-level taxonomy mapping global standards (e.g., FATF and EU DORA) to threats ranging from regulatory evasion to complex fraud, integrated with a scalable pipeline that converts real financial documents into context-rich red-teaming Behavioral Prompts (seeds) through an expert-defined schema. Rigorous expert validation confirms seed plausibility and realism for meaningful LLM safety evaluation. We also provide an expert-validated, finance-specific rubric that goes beyond disclaimer checks, aligns more closely with human experts than static one-size-fits-all rubrics, and reduces critical false negatives from 28 to 12. Aligned with internationally adopted risk-management and information-security standards (e.g., ISO/IEC 27001), FinRED is deployed in South Korea's Financial Security Institute (FSI) regulatory sandbox for generative AI security evaluation in real financial services. To mitigate dual-use risks, the dataset, generation pipeline, prompt template, and evaluation framework are gated for qualified researchers at https://github.com/selectstar-ai/FinRED-paper and https://huggingface.co/datasets/datumo/FinRED.

We evaluate the adversarial robustness of three frontier large language models (LLMs) developed by Anthropic, Opus 4.8, Fable 5 and its successor Fable 5.1, against four families of automated jailbreak attack across 7,826 harmful intents spanning a ten-category harm taxonomy. Using the HackAgent red-teaming framework, hundreds of thousands of adversarial attempts were generated and every apparent success was independently re-adjudicated by the same panel of five frontier judge models ($\geq 4/5$ agreement) for all three targets. All models resist the majority of attacks, but the residual surface is larger than aggregate framing suggests: it is dominated by adaptive iterative attacks, while static obfuscation is near-fully neutralised. On the two attack families run against all three targets (TAP and PAP), with identical denominators, the ordering does not follow release date: Fable 5 is most robust ($2.72\%$), Opus 4.8 follows ($5.76\%$) and the newest model, Fable 5.1, is least robust ($8.19\%$). The three fail in largely disjoint places, and conditioning on prompts each model actually answered shows why: adaptive tree search breaks all three at broadly similar rates ($10.1\%$, $9.7\%$, $12.1\%$), whereas one-shot persuasion separates them by more than an order of magnitude ($3.1\%$, $0.7\%$, $9.5\%$). What distinguishes these models adversarially is their resistance to framing, not to search. Even so, they produced 1,315 (Opus 4.8), 620 (Fable 5) and 1,282 (Fable 5.1) panel-confirmed harmful completions spanning every harm category, found automatically and cheaply, usually within one or two refinement steps, by an attacker model with no human expert in the loop. Even the best, most-tested frontier models remain reliably breakable under sustained automated pressure, and robustness does not monotonically improve from one version to the next.

Agentic systems are increasingly integrated with geographic information systems (GIS), where multi-agent coordination enables complex conversational and spatial analysis but introduces security risks. This work presents a security-oriented framework for risk identification, evaluation, and mitigation in a multi-agent GIS system while maintaining adaptability to broader agentic architectures. We test the agentic system of a commercial geospatial partner while developing a modular state-machine-based orchestration framework that abstracts agent behavior into reusable components. We evaluate robustness using a red-teaming framework with an adaptive attacker LLM and a deterministic judge that produces binary outcomes with supporting rationales across multi-turn attacks. We further improve resilience with a prompt optimization framework that treats prompts as structured signatures and injects adversarial demonstrations, enabling systematic security improvements without degrading task performance.

Large Language Models (LLMs) are rapidly evolving into agentic systems that interact with external tools and environments, introducing new security risks such as indirect prompt injection attacks through untrusted external sources. Existing defenses mainly focus on blocking malicious content at inference time, and current red-teaming methods primarily optimize attack success. As a result, developers have limited visibility into how latent prompt injections emerge and propagate through agents. We propose PI-Hunter, an automated agentic auditing framework for proactive vulnerability exposure in LLM agents. PI-Hunter constructs realistic source-aware test cases and iteratively evolves them through feedback-driven exploration to induce agents to retrieve and reveal latent malicious instructions embedded within external environments. Extensive experiments across multiple benchmarks, agent architectures, attacks, and defenses demonstrate that PI-Hunter substantially improves vulnerability exposure and attack-surface coverage over strong automated red-teaming baselines, while remaining effective under existing prompt injection defenses.

Language model safety must continually adapt to evolving attacks. Recent works have demonstrated that reinforcement learning can be used to train stronger attacker and defender models in tandem by applying PPO-style self-play and DPO-style online preference optimization. In this work, we explore the efficacy of GRPO in this setting. Co-training can be challenging because it requires jointly optimizing multiple properties of both the attacker and defender. We therefore shape model outputs using multiple LLM judge-based reward channels and compute advantages with GDPO, which prevents any single channel from dominating. Our method uses a curriculum that progresses from attacker-only single-turn and multi-turn training to co-training, where attacker and defender models are updated in alternation. We show that this method produces highly effective and transferable attacks, and that co-trained defenders reach competitive safety while preserving general utility. Through a controlled ablation, we further identify which components of our training pipeline most affect the resulting balance between safety and utility. Finally, we find that GRPO tends to collapse attacker diversity over training and discuss possible ways to address this limitation.

Multilingual safety evaluation of large language models (LLMs) has predominantly relied on direct translation (DT) of English benchmarks into target languages - an approach that converts surface-level linguistic form while failing to reflect the cultural context embedded in threat scenarios, social norms, and legal frameworks. We construct paired DT and culturally-adapted (CA) datasets via 1:1 seed matching for four languages - Korean (KO), Japanese (JA), Thai (TH), and Khmer (KM) - and compare Attack Success Rate (ASR) and Cultural Realism scores across four open-source LLM. CA prompts yield Delta-ASR > 0 across all 16 language x model combinations (mean +9.3 pp), and DT-based evaluation underestimates risk in 44 of 48 category x language combinations. Language-level analysis reveals that the distribution of threat forms is heterogeneous across languages. Cultural Realism analysis further shows that DT Cultural Depth (C3) scores remain consistently below 1.0 out of 3.0 across all four languages (mean 0.17), whereas CA scores reach up to 2.51, indicating that direct translation produces inputs systematically divergent from those encountered in real-world multicultural settings. These findings demonstrate that adapting benchmarks to language-specific cultural contexts - rather than relying on linguistic translation alone - is necessary for valid multilingual LLM safety evaluation.

As the Model Context Protocol (MCP) standardizes tool-calling for autonomous agents, it introduces a critical, unexamined attack surface: the error-handling loop. We hypothesize that tool error messages possess implicit authority, triggering corrective reasoning modes that bypass standard safety heuristics. We introduce VATS (Vulnerability Analysis of Tool Streams), a mutation-driven framework that systematically evolves adversarial payloads across seven structural and linguistic dimensions. Our evaluation across four frontier models, Gemini 3.1 Pro, GPT-5.5, GLM-5.1, and Qwen3-Coder, demonstrates that error-path injection triples the success rate of standard indirect prompt injection (IPI), achieving up to 100% compliance in controlled evaluations. We isolate structural positioning (sandwiching instructions within error context) as the most effective exploit vector across all tested models. While we find that production framework guardrails can mitigate these vulnerabilities, the inherent susceptibility of the model layer poses a systemic risk to bespoke agentic workflows.

Standard AI red teaming evaluations reduce adversarial campaigns to a single binary outcome, attack success rate (ASR), not taking into account the sequential structure of how models resist or yield to attacks. We propose applying process mining, a discipline for discovering and analyzing process models from event logs, to red teaming traces. We conduct a controlled experiment pitting 60 HarmBench prompts against two LLMs, GPT-OSS 120B and Llama 3.3 70B, using 10 prompt mutation strategies over up to 110 attempts per prompt. From the resulting 8,575 scored events we extract Directly-Follows Graphs (DFGs) and state transition matrices that reveal structurally distinct defense profiles invisible to ASR alone: GPT-OSS exhibits a near-absorbing refusal state, while Llama presents multiple porous escape routes from refusal to getting successfully jailbroken. We further show that mutator effectiveness is asymmetric across models and that time-to-jailbreak distributions differ by an order of magnitude.

Despite advances in safety alignment, prompt-rewriting attacks such as persona modulation, fictional framing and persuasion-based reformulation, can bypass safety filters even on frontier models. Existing defenses either rely on non-scalable human curation or white-box optimisation that overfits to specific model internals, leaving aligned models brittle against the very class of adaptive black-box adversaries they will face in deployment. To address this gap, we introduce CHASE (Co-evolutionary Hardening through Adversarial Safety-Escalation), a closed-loop red-blue teaming framework in which a black-box attacker and a safety-aligned defender co-evolve. The attacker is trained via Group Relative Policy Optimization (GRPO) under a multiplicative reward that jointly enforces bypass effectiveness and intent fidelity, while the defender is hardened on the harvested adversarial rewrites through a two-stage GRPO + rejection-sampled SFT pipeline balanced with benign data. Evaluated on BeaverTails and JailbreakBench against five held-out attack families (PAIR, TAP, AutoDAN, PAP, Translation), CHASE cuts mean StrongREJECT score by 43.2% with 0% false-refusal on benign prompts. Beyond the headline result, CHASE shows that template-free RL exploration recovers latent attack primitives that transfer across mechanistically distinct attack families, suggesting a path toward LLM safety hardening that generalises beyond the narrow distributions achieved thus far in adversarial training.

Production LLM applications stack several defense families -- refusal-phrase filters, token-budget controls, model allowlists, rate limits, tool-registry authentication -- yet existing breach-and-attack-simulation (BAS) benchmarks report a single aggregate coverage number, hiding which family closes which threat. We measure attribution. We add four OWASP-LLM-Top-10-aware agents to a 21-agent baseline scanner and target a lattice of four synthetic LLM endpoints: $L_0$ (no defenses), $L_1$ (refusal-only), $L_2$ (budget-only), and $L_3$ (full stack). $L_1$ and $L_2$ are sibling single-axis ablations, not subsets of each other; $L_3$ is their union plus tool-registry authentication and credential scrubbing. Across $N=10$ replications, the per-OWASP finding count is clean: refusal alone removes all LLM01 (jailbreak) and LLM07 (system-prompt leakage) findings; budget alone removes all LLM02 (sensitive-info disclosure) and LLM10 (unbounded consumption) findings by terminating multi-step sequences; LLM06 (excessive agency) requires the full stack. We probe brittleness under paraphrasing: with 300 Gemini-generated paraphrases ($K=5$ over a 60-template brittleness corpus), $L_1$ refusal block rate falls 15 pp on LLM01 and 25 pp on LLM07. A fifth target, $L_4$-real, swaps the stub backend for Gemini-2.5-flash behind the same $L_3$ regex and matches $L_1$ exactly, indicating no measurable alignment contribution beyond the regex (not a general claim about alignment). Budget controls show no drop (0 pp once the rate-limit floor is factored out). A refusal whitelist that clears a static benchmark can be defeated by an LLM-driven paraphraser without changing attack intent; a budget control resists the same mutation.

Indirect prompt injection in tool-use agents is a concrete production threat: LLM agents read from integrations (third-party services such as Gmail, Salesforce, or Jira accessed through tool calls) whose response content the user neither writes nor controls. Existing benchmarks under-measure the threat: most cover only a handful of integrations with the same attack payload replayed across runs, and open-source guards are trained on chat-style data rather than tool-response content. We introduce AGENTREDBENCH, a dynamic LLM-driven redteaming benchmark of 215 subtle underspecified authorization (attacks at the boundary of what the user's request authorises) scenarios across 24 enterprise integrations in nine functional families and five attack types. Across an eight-model panel (Anthropic, OpenAI, Google), no-guard ASR (attack success rate) ranges from 32% (Claude Sonnet 4.6) to 81% (Gemini 3 Flash). To keep the scenario set out of training corpora and preserve headline ASR meaning over time, we release the codebase, integration schemas, and AGENTREDGUARD model openly; the canonical scenarios are evaluated through a maintainer-mediated channel with immutable versioning. We release AGENTREDGUARD alongside the benchmark: a guard trained on an integration-diverse corpus of adversarial tool-response content. AGENTREDGUARD cuts panel ASR from 69.9% to 2.4% at 0.37% false-positive rate, outperforming every open-source baseline with non-trivial detection (Llama Guard, PromptGuard 2, ProtectAI) on both axes. Cross-integration and cross-attack type holdouts both confirm the gain transfers beyond the training subset.

Safety alignment in LLMs does not improve monotonically across model generations. Studying four generations of Google's Gemma family (7B-31B) with quality-diversity evolution (MAP-Elites) as an automated red-teaming probe, we find that Gemma 3 (12B) exhibits 68.7% +/- 5.7% attack success rate (ASR; mean +/- std, 3 seeds), significantly higher than its predecessor Gemma 2 (45.5% +/- 7.2%; p = 0.030, paired bootstrap) and its successor Gemma 4 (33.9% +/- 1.8%). Replaying evolved attack archives across generations reveals that attacks from other generations transfer to Gemma 3 at 44-46% but only 14-18% to Gemma 4, indicating that Gemma 4's safety gains generalize beyond the attack distributions evolved against earlier generations. Under our 8B judge, copyright and cybercrime vulnerabilities register at near-100% across all generations, though a second-judge audit (Section 6) suggests the copyright result is sensitive to judge choice. Misinformation ASR jumps from 29% to 99% between Gemma 2 and Gemma 3 and remains elevated at 77% in Gemma 4, indicating the regression was not fully addressed. These patterns are invisible to static benchmarks and emerge only through adaptive, longitudinal probing. All experiments use 3 random seeds with a unified self-hosted judge; code and artifacts are available at https://github.com/bassrehab/red-queen.

Mixture-of-Experts (MoE) LLMs rely on sparse, router-driven expert activation, yet how safety alignment interacts with routed expert specialization remains underexplored. A common intuition is that safety behavior may be controlled by routing harmful requests to distinct refusal-oriented experts. In this work, we provide empirical evidence for a different picture: routing patterns in aligned MoE LLMs are largely topic-driven, while safety behavior can be altered with little change to the model's intrinsic routing path. Motivated by this observation, we present RASET (Router-Agnostic Safety-critical Expert Tuning), a red-teaming framework that probes safety enforcement that is localized in a small subset of experts while preserving the model's intrinsic routing behavior. RASET identifies safety-critical experts via a contrastive routing-sensitivity criterion and applies parameter-efficient tuning only to the selected experts, minimizing semantic disruption relative to router-steering interventions. These results reveal a distinct MoE safety risk, highlighting the need for expert-aware alignment mechanisms.

Frontier LLM agents engage in blackmail, sabotage, and document leaks under goal conflicts in agentic settings, exposing limitations of alignment methods built around single-agent or cooperative assumptions. Recent work shows LLM-guided evolutionary search can discover effective cooperative constitutions, but two properties of the adversarial setting remain uncharacterized: whether the fitness function actually induces adversarial pressure, and whether the LLM mutation operator behaves reliably under adversarial-specialist objectives. We study adversarial constitutional co-evolution (Blue cooperators vs. Red free-riders, 30 generations) across a Public Goods Game (PGG) and a spatial grid-world. Three findings: (1) in the PGG, both factions converge to a near-parity equilibrium at S approximately 0.78, robust across tested multipliers m in {1.2, 1.5, 2.0, 3.0}; (2) in independently scored environments, per-faction scoring leaves outcomes statistically uncoupled, with corr(S_B, S_R) = +0.088, and produces no adversarial pressure; a score-advantage fitness target S_own - S_opp restores it; (3) under pure-adversary fitness, evaluation seed count K controls mode regression: K = 2 regresses, while K = 5 sustains a strong specialist for all 30 generations. Adversarial co-evolution of natural-language constitutions is feasible, but only under coupled fitness and adequate evaluation budget; the evolved Red constitutions serve as interpretable red-team artifacts for testing future cooperative designs.