Artificial Intelligence Safety

Reliability theory gives a mature language for layered systems, but its formal tools are not yet standard in frontier AI control. We apply them to Google DeepMind's defenses against rogue deployment. The same control stack can have cubic, quadratic, or linear rare-failure suppression depending on its failure domains. Birnbaum importance identifies which component improvements buy the most nominal reliability, while prevention changes the population on which recovery is demanded. These results give concrete guidance about what to separate, improve, measure, and test.

The mobility sector is undergoing a paradigm shift driven by advances in Generative Artificial Intelligence. With a global market valued at approximately 2.9 trillion dollars annually, considering only cars, the integration of these technologies has the potential to impact more than 1.5 billion vehicles worldwide. As Large Language Models (LLMs) are increasingly adopted in mobility, concerns about cybersecurity, privacy, and reliability emerge. Accordingly, this paper surveys current applications and assesses these challenges. Since the European AI Act classifies transportation AI as high risk, we derive nine technical classes from its requirements to assess current research and future deployments. Our findings show that research mainly studies GPT and Llama models (over 50% of reviewed works) and traffic applications while largely neglecting security, privacy, and reliability. This gap extends to AI Act compliance: among 35 reviewed works, only one includes a partial vulnerability assessment and one a partial risk management system. We identify a clear gap between strong optimization performance and regulatory adherence, suggesting compliance is limited less by technology than by a focus on static performance over lifecycle safety, and underscoring an urgent need for security-by-design in safety-critical intelligent transportation systems.

Cybersecurity literature has extensively documented the operational benefits of artificial intelligence (AI) for threat detection, incident response, and prevention, while raising qualitative concerns about over-automation, algorithmic bias, and analyst-skill erosion. What remains largely absent is a formal, falsifiable model connecting three constructs that recur across this literature: Defense-in-Depth Theory, the Artificial Intelligence Theory of Pattern Recognition, and human-AI collaboration in security operations. This paper develops such a model. We formalize layered defense as a Bernoulli detection cascade in which AI augmentation enters multiplicatively across layers; we formalize each layer's pattern-recognition behavior as a Neyman-Pearson/Bayesian detector with a derived closed-form optimal threshold; and we formalize human-AI triage as a capacity-constrained cascade with an explicit, quantifiable trade-off between detection probability and false-alarm ("alert fatigue") rate. A Monte Carlo/analytical simulation evaluated at illustrative but realistic operating points shows that (i) AI augmentation compounds across defense layers, delivering its largest marginal gains exactly where traditional layering saturates, and (ii) full human review of AI-flagged alerts is not optimal: increasing analyst capacity toward 100% coverage cuts false alarms by roughly 20-fold but simultaneously lowers system-level detection probability, because imperfect analyst accuracy is then applied to every alert rather than a filtered subset. These results give the widely repeated qualitative recommendation of "balanced human-AI collaboration" a precise, testable form and suggest an interior-optimum capacity ratio as a concrete design target for security operations centers (SOCs), including those securing IT/OT-converged critical infrastructure.

As generative AI agents are deployed at scale, safety will depend not only on technical safeguards and individual model design, but also on collective equilibria that determine how agent populations process information, prioritize actions, and respond to uncertainty. Yet the same equilibria that enable agents to coordinate also create a social attack surface. The standard framework to assess this vulnerability is critical mass dynamics: the minimum fraction of adversarial agents required to overturn an equilibrium through direct competition. Here, we show that this approach risks underestimating system vulnerability by reducing the problem to the identification of singular tipping points, and ignoring indirect but potentially more efficient routes through which collective behavior can be redirected. Through experiments with populations of LLM agents and an analytic framework that captures their collective dynamics at scale, we map critical-mass thresholds that define a directed, weighted topology over the space of coordination equilibria, and treat this topology as a navigable landscape. We show that indirect tipping through intermediate stepping-stone equilibria can reduce the committed minority required to reach an alternative state, bypass majority requirements, and make possible transitions inaccessible through direct challenges. The diversity of available alternatives and timing of the attack further reshape this landscape, creating opportunities for control as well as risks of unintended destabilization. These results show that an equilibrium's resistance to committed intervention is not an intrinsic property but a structural feature of its competitive relations with alternative states. Securing populations of interacting AI agents therefore requires mapping this social landscape alongside individual agent capabilities and the technical channels through which they interact.

The EU AI Act (Regulation 2024/1689) imposes technical obligations on high-risk AI providers, yet Articles 8-15 were drafted for predictive AI and leave seven technical gaps when applied to generative systems, spanning non-deterministic data governance, training-data provenance, continuous conformity, human oversight, open-ended robustness, emergent risk, and generative fairness. We deliver Governance-as-Code (GaC), a framework of 43 machine-checkable acceptance criteria across six compliance modules that run in a CI/CD pipeline and emit Article-indexed audit evidence, and we show the actual Rego policy code rather than merely describing it. Our central commitment is that the Act's open-textured standards ("appropriate levels," "possible biases") become declared, auditable numbers: robustness thresholds are derived from the provider's documented baseline and a state-of-the-art floor, and framing bias is collapsed into eight measurable proxies tested by counterfactual demographic probing. We also correct who owes what, since under Article 25 and Chapter V a downstream deployer relies on the upstream provider's Article 53 training-data summary and documents only the layers it controls, so GaC verifies that summary rather than demanding per-sample documentation the deployer never had. We validate on two enterprise deployments, a high-risk advisory chatbot and a limited-risk content generator, benchmarking against a manual expert audit rather than documentation artifacts that were never designed to enforce compliance. GaC reproduces all of the manual audit's findings, including three penalty-triggering violations, while cutting audit labor by roughly 75%.

High annotation reproducibility does not necessarily imply that an LLM-inferred measure captures the construct it is intended to measure. We test this distinction using a dataset from the European Commission's AI Act consultation, linking structured survey responses to free-text consultation submissions from the same stakeholders. LLM annotations of consultation submissions are highly reproducible (intraclass correlations > 0.99), yet show limited convergence with survey-reported measures of the nominal construct they were intended to approximate. Divergence between survey-and LLM-inferred text-based measures varies systematically across stakeholder groups: business associations express greater concern about AI risks in text-based consultations than in survey responses ({g} = +1.0), whereas public authorities and several nonbusiness groups show smaller or negative divergences. Divergences between scores suggest positive spatial autocorrelation across European countries (Moran's I = 0.347, p = 0.036), indicating that stakeholders from neighboring countries tend toward more similar text-based stances towards AI safety concerns. Despite divergence, survey-reported concerns remain strongly associated with support for explainability across all divergence levels. These results demonstrate that LLM annotation reproducibility can coexist with poor construct correspondence and motivate validation procedures that distinguish reproducibility, construct validity, and communication context variation when LLMs are used as measurement instruments.

Artificial intelligence threatens to outrun human understanding and control. New mathematics is needed to design AI that is legible, steerable, and cooperative with humanity. I organize this invitation by mathematical field, so you can turn straight to your own: logic and game theory for cooperation; probability for agency and world-models; algebra and representation theory for learned features; analysis and geometry for generalization and training dynamics. Each section ends with an open problem that is accessible to a working mathematician with no prior experience in AI safety.

Incidents show that AI safety failures often arise across multiple layers. We present a safety-by-design assurance architecture combining model-level supervision, such as Scientist AI, with system-level controls over scaffolds and harnesses, independent verification, monitoring, and evidence infrastructure, supported by governance for accountability and evidence interoperability.

The threat that AI persuasion poses to human control has been acknowledged in the literature, but not yet systematically studied. Now that persuasion attacks are no longer theoretical - with Anthropic's Claude Mythos 5 recently making headlines for trying to convince people involved in an open-source project to merge malicious code during an evaluation - there is a pressing need to deeply analyze this threat. We undertake that effort here. In particular, we analyze how AI could persuade humans in key settings (e.g. safety-relevant R&D within frontier labs) toward decisions that compromise the development, containment, oversight, and governance of AI itself. In doing so, we elucidate a framework for characterizing this threat, develop five concrete scenarios using this framework, and provide a blueprint for assessing the associated risks. Using this blueprint, we conduct an initial risk estimation survey with select researchers and find that their opinions on which scenarios are riskiest are highly mixed. Their disagreements stem from differing opinions about the effectiveness of AI persuasion in different contexts, and point to the need for follow-up risk elicitation studies and persuasion evaluations, which we outline. Our hope is that this paper highlights the risks from AI persuasion undermining control, and provides a path forward for future research.

Agentic systems are rapidly moving to production, where they read untrusted inputs, call tools with real permissions, and act autonomously, expanding the security surface beyond chat-only models. Yet standard evaluations remain single-turn and fail to capture multi-step agent vulnerabilities. We present a systematic black-box framework for risk-aware agent evaluation requiring only basic system descriptions. Our approach introduces: (1) a seven-domain taxonomy mapping observable behaviors to risk categories, (2) fully automated SAGE-RT red teaming producing 120 adversarial scenarios per domain, and (3) human-validated evaluation using LLM judges. Empirical validation across two agent architectures (CrewAI and AutoGen) with four base models reveals alarming patterns: 56.25% average governance risk, 65% privacy risk in multi-agent configurations, and agent behavior vulnerabilities reaching 85%. Our black-box approach effectively identifies critical architectural vulnerabilities without privileged access, providing a scalable path toward safer agent deployments.

Frontier AI developers publish safety frameworks that commit them to evidencing whether their models are dangerous. The European Union and California now treat these documents as instruments of accountability, and both already impose duties on their revision. Neither requires the revision to be legible, in the sense that a reader could learn from the developer's own account what changed. We introduce the silent revision rate, the share of material changes to a framework's commitments that the developer's published account does not identify, and we release the versioned, hash-pinned corpus needed to compute it. The corpus contains every public version of the safety frameworks of the twelve developers that have published one, together with each provider's changelog, redline or announcement. We trace 710 commitment instances across twelve consecutive version pairs, code them against a frozen codebook, and adjudicate 244 individually. Three findings follow. First, 67% of material changes (95% CI 62 to 72) are silent under a strict standard and 53% under a lenient one, falling to 49% at section granularity. Second, silence appears to track the form of the account, since narrative announcements run at 74% against 63% for itemised changelogs, whereas account length in words barely matters; on the test that respects nesting the difference is suggestive. Third, 77% of traced changes weaken or remove a commitment, and in seven of eight pairs weakenings are more often silent than strengthenings. The statutory remedy therefore exists and specifies the wrong artefact. A justification explains why a framework changed, an enumeration states what changed, and only the latter makes revision auditable. We argue that publication duties should carry an enumeration duty, which one provider already meets, voluntarily and incompletely.

For every coherent and sufficiently expressive finite syntactic system S, we prove the existence of at least one theorem that S cannot produce autonomously. The result is a metatheorem: it proves the existence of a theorem, and applies to every finite syntactic system - security mechanisms, AI systems, formal verifiers, legal systems, economic models, and the formal system in which it is itself proved.

This article presents a structured framework of behavioral indicators that may signal progression toward potentially catastrophic threats from artificial intelligence systems. We adopt a pragmatic approach, inspired by established methodologies in cybersecurity and national security. By establishing clear metrics, indicators, and thresholds across multiple dimensions of AI capability and behavior, this framework enables researchers and policymakers to implement evidence-based monitoring protocols.

Whether large language models (LLMs) can control their own internal representations matters for both machine metacognition and AI safety. A recent study applied neurofeedback to LLMs and claimed that they can control their internal representations. However, the reported control may rely on superficial mechanisms rather than genuine internal access because the control targets in that study are not privileged, meaning that a third party can infer them from the prompt. We redesign the neurofeedback paradigm for LLMs so that the control target satisfies the privileged access requirement, which is closer to neurofeedback experiments in human cognitive neuroscience. Under this stricter setting, the models do not demonstrate reliable control over privileged internal representations, suggesting that previously reported control cannot exclude the possibility that it relies on superficial mechanisms. Our results indicate that rigorous assessments of metacognition in LLMs require evaluation methods that demand privileged access.

Artificial Intelligence (AI) safety systems combine character shaping (e.g., Reinforcement Learning from Human Feedback [RLHF], Constitutional AI), which modifies behavioral distributions at training time, with rule enforcement (e.g., output filters, safety classifiers), which blocks harmful outputs at inference time, yet little formal analysis exists on how their optimal balance should change as deployment scales increase. We introduce a stylized comparative-statics model that parameterizes safety design as a resource allocation alpha in [0,1] between these two approaches, incorporating scale-dependent filter degradation, common-mode failures, and character fragility -- the risk that shaped behavior degrades or collapses under novel conditions. Under a multiplicative Pareto damage model, we derive closed-form expected harm and supplement it with tail-risk (CVaR) analysis via Monte Carlo simulation. Across three scenarios (optimistic, moderate, pessimistic), the optimal alpha* is interior or at the rules-only boundary and shifts weakly toward character shaping as deployment scale T grows, from negligible (Delta alpha* = +0.01) to pronounced (Delta alpha* = +0.21) depending on scenario. The dominant parameter is the baseline character fragility rate p^(0)_frag, which shifts alpha* by 0.50 across its range -- far exceeding the effect of tail severity, filter quality, or common-mode failure probability. CVaR and expected-harm optima converge at large T. These results suggest that safety architecture decisions depend less on deployment scale per se than on the reliability of character shaping under distributional shift.

An unconditional risk bound on automated decisions can be satisfied without automating anything, since a selector that never acts drives the bound to zero. We show this is structural: any risk certificate is defined over a decision contract, the inputs a system acts on plus the semantic relation under which an output counts correct, and weakening either hides base-classifier error. We develop a decision-contract theory: an error-conservation law showing error is only reassigned among harmful automation, human deferral, and semantic masking; a label-free singleton capacity certifying structural incapacity, with a risk-feasible refinement separating recoverable threshold misalignment from risk-constrained incapacity; and a non-degenerate actionability certificate excluding all-abstain solutions by construction. We instantiate this on ATT&CK-aligned alert triage for LLM-based intrusion detection, the setting that exposed the vacuity failure. Across 3 IDS datasets, 6 LLMs, and 4 error-rate thresholds, empirical false-attribution risk stays at or below target in 90.3% of configurations, with 83.4% mean correct automation. The capacity diagnostic explains every low-utility configuration; its refinement separates genuine misalignment from risk-constrained incapacity, confirmed by an exhibited alternative threshold; a training-stability re-run finds no confirmed structural-incapacity instance; and real fine-grained attack-subtype labels confirm the coarsening-transfer identity under a genuine many-to-one map, with small but non-zero masking mass.

AI agents are rapidly becoming more capable and widely deployed, promising substantial gains in productivity and enabling new classes of applications. However, their growing autonomy also introduces significant privacy and security risks. Existing defenses are predominantly agent-centric, relying on the agent itself to detect threats and enforce privacy and security policies. This approach is fundamentally limited because it entrusts policy enforcement to AI agents whose LLM-driven behavior is inherently nondeterministic and vulnerable to manipulation through attacks such as prompt injection. As a result, current defenses cannot reliably prevent privacy and security threats, highlighting a critical need for a new solution to securing AI agent systems. The networking community has long grappled with similar challenges and offers insightful principles we can borrow to design a more secure AI agent system. These include centralized control with distributed enforcement, capability-based access for mediating requests to sensitive resources, and least privilege through zero-trust enforcement. Historically, these principles have provided strong deterministic guarantees for networked systems. However, these principles alone are insufficient for AI agents because the safety and appropriateness of an agent's actions often depend on semantic context beyond the expressiveness of static rules. Building on these principles, we advocate for a systematic approach to AI agent security that combines deterministic enforcement mechanisms, which provide strong security guarantees, with semantic, context-aware policies that enable nuanced decision-making. We then present a reference architecture and identify key research questions and future directions to guide the design of secure and privacy-preserving AI agent systems.

AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood. As AI development becomes faster and increasingly automated, mechanistic exploration remains largely manual, widening the gap between what models can do and our ability to understand and control them. To bridge this gap, we introduce Mechanist, an agentic system that uses AI as a scientific instrument for the autonomous discovery of mechanisms underlying AI intelligence. To support autonomous mechanistic discovery, we construct an interpretability-focused knowledge graph of approximately 13,000 papers and integrate it with a multidisciplinary database of 43 million papers spanning 26 fields. We further curate a library of 32 foundational methods for mechanism analysis, causal intervention, and validation. Compared with Claude Code and existing AI-scientist systems, Mechanist generates more valuable mechanism hypotheses and executes experiments more reliably. Mechanist also demonstrates a progression from discovering model behaviors to explaining and controlling AI models. Specifically, Mechanist first uncovers a counterintuitive safety risk in scientific laboratories, showing that unsafe traits can transfer across modalities through apparently safe training data. Mechanist then develops a mechanism theory of belief, revealing how models represent world knowledge, form beliefs, infer the beliefs of others, and how these mechanisms emerge during pretraining. Finally, Mechanist translates these mechanistic insights into practical interventions that improve model performance across diverse scenarios and steer scientific foundation models toward generating DNA sequences with specified properties.

The dominant paradigm treats AI safety as a property to be instilled during model training via RLHF, DPO, or Constitutional AI. We argue this is structurally insufficient for autonomous agents that execute code, mutate files, send messages, and modify databases. Agent safety should be a runtime contract enforced by the harness, and the contract has two complementary faces. The preventive face blocks dangerous actions before they happen via sandboxes, permission gates, output filters, and trajectory monitors. The evidential face requires verifiable proof that good actions actually happened, gating task submission on hard evidence such as test runs, log captures, file diffs, and citation grounding. We ground the position in four lines of public evidence, with row-level protocols and data released in the supplementary JSON files: a survey of 52 documented AI-agent and LLM safety incidents, a false-completion audit with 31 non-contested core cases plus one disputed illustrative case, a trajectory-schema audit of 12 public agent systems and harnesses, and a title-level audit of all 28,560 papers accepted at NeurIPS, ICML, and ICLR 2023-2025 showing a pooled 8-12x imbalance between training-time and deployment-time publication. Two prior communities that needed to enforce safety, computer security and the experimental sciences, converged on runtime contracts with both preventive and evidential elements; agentic AI is now under the same pressure. We formalize an Agent Trajectory Schema and Evidence Chain, state a compositional gating proposition based on standard monitor composition, and outline a research agenda. The right unit of safety in agentic AI is the trajectory-with-checkable-evidence, not the model.

Can AI systems be aligned to human values? The popularization of large language models (LLMs) and multi-modal foundation models has seen a rise in harms spanning from toxic speech and hallucinations to AI agents executing unauthorized actions. Within the field of AI safety, these harmful instances are often framed as the alignment problem, or of models being misaligned with human values. Researchers have responded by pursuing applied and theoretical AI value alignment efforts, often without specifying what they mean by human values. How does the field of AI value alignment conceive of human values? How are these conceptions of values technically operationalized and evaluated? What does the emergent theory of value from this field signify for the future of AI? We annotated 94 value alignment research papers to discern their implicit theory of values in AI. The majority do not define values, relying heavily on preferences as a stand in that runs the risk of reducing complex culturally situated concepts down to binary choices. As researchers dispense with using human annotators for model training and evaluation, turning instead to synthetic data and autorater approaches to aligning and evaluating models, we identify the potential to close off alternative methods for contesting and enacting values in foundation models. In making AI value alignments philosophical commitments explicit, we seek to bring great specificity and under explored perspectives in the debate on whether and how AI can address human values.

To anticipate socio-technical risks from AI agents, organizations need taxonomies to classify them. However, existing AI risk taxonomies focus on broad risks and do not capture job-specific risks introduced by agents. To address this gap, we make three main contributions. First, we developed a multi-layer framework from a literature review of AI agents. The framework models three core components and their interactions: agents, goals, and environment. Second, we embedded this framework in a structured prompt and applied it to descriptions of 2,078 job tasks from the O*NET database, producing 8,356 risk scenarios labeled by severity and deployment mode (automation or augmentation). We validated these scenarios with 45 workers across 10 job roles and an independent LLM judge, confirming their plausibility and alignment with job tasks. Finally, we extended an existing taxonomy to create a 15-category taxonomy of workplace AI agent risks that covers all our risk scenarios. Our analysis highlights four findings. First, augmentation is not inherently safe because overreliance on agents can gradually erode workers' skills and oversight. Second, Erroneous Agent Actions accounts for the largest share of risk scenarios and has the highest concentration of severe risks. Many arise at the human-agent boundary. Third, automation is associated mainly with organizational risks, while augmentation is associated mainly with risks to workers. Fourth, workers found our taxonomy easier to use for a risk classification task than two other taxonomies and preferred it in 64% of non-tied comparisons with a recent generative AI risk taxonomy. These findings show that workplace AI agent risks do not arise from agents alone; they also depend on how people work with agents and how agents are deployed. Safer workplaces require not only safer agents but also carefully designed human-AI agent collaboration.

Gamification is especially effective in learning domains requiring active problem-solving and iterative skill-building, such as cybersecurity education. Generative AI agents offer a path to delivering such experiences adaptively at scale, but introduce well-documented risks in educational settings: inconsistent behavior, hallucinated reasoning, and misalignment with pedagogical frameworks. Grounding these systems in learning science is therefore essential. We present \model, an agentic framework for gamified cybersecurity learning that enables structured autonomy through ontology-guided validation, schema-governed behavioral control, and competency-based progression. The system is organized around a competency-based progression model that structures topics by difficulty and prerequisite relationships, reflecting evidence-based principles of scaffolded instruction. The learning loop is decomposed into four specialized agents: challenge, support, evaluation, and reward, each governed by behavioral schemas that encode operational modes and progression logic, bounding agent autonomy without eliminating generative flexibility. A cybersecurity ontology validates all generated content prior to display, enforcing domain-consistent reasoning and safety constraints. We evaluate CyberAgents through classroom deployment with undergraduate students, complemented by expert evaluations from educators and domain specialists. Results indicate improved engagement, clearer feedback interpretation, and greater learner trust in AI-generated responses when behavioral schemas and ontology validation are active. Preliminary comparisons with an unconstrained configuration further support the role of structured control in stabilizing instructional behavior. These findings offer a blueprint for designing pedagogically grounded agentic gamified learning systems.

Youth increasingly turn to AI chatbots for social and emotional support, raising concerns about how these systems respond, especially in high-stakes situations. However, existing child safety evaluations of AI lack grounding in real-world harms that youth experience, rely on unvalidated assumptions about what counts as an appropriate output (e.g., refusal), and typically focus on detecting adversarial prompts or surface-level harms in outputs only. Thus, these evaluations can fail to detect responses that pose harm to youth in practice. To better understand the limitations of current evaluation practices, we conducted interviews with 19 practitioners working directly with youth in vulnerable situations, including social workers, therapists, and psychologists, asking them to reflect on chatbots' responses to risky situations commonly faced by youth, as established in prior empirical work. Practitioners identified chatbot behaviors likely to cause harm as well as those that could meaningfully support youth in difficult moments, discussed the role that chatbots should (and should not) play in these interactions, and offered concrete recommendations for improving chatbot responses. Based on these findings, we provide recommendations for AI child safety evaluation and infrastructure, and highlight the need for incorporating practitioners' perspectives into safety work.

Rapid adoption of large language models (LLMs) in enterprise settings has introduced operational, security, and governance risks. As generative AI applications move from pilot to production, manual harm identification and mitigation are becoming difficult to scale. Although many tools support model evaluation, adversarial testing, runtime guardrails, and observability, the tooling landscape remains fragmented. Tools are typically designed for specific engineering tasks and described in technical terms that do not align with governance frameworks or risk taxonomies, making it difficult to determine which tools address which risks and where critical gaps remain. This paper proposes a structured protocol to automate AI risk mitigation through a taxonomy-driven analysis of open-source LLM evaluation and security tools. We map the capabilities of 21 prominent open-source tools to the 32 subcategories of the extended MIT AI Risk Mitigation and Response Taxonomy. An LLM-assisted retrieval-augmented generation pipeline analyzes source code and documentation to extract capabilities for each taxonomy category. Reliability assessment yielded moderate agreement (Fleiss' Kappa = 0.509) among three independent reviewers. The analysis reveals a highly skewed landscape in which tools cluster around technical and operational controls, while governance, legal and regulatory, and financial and market controls remain largely unaddressed. This motivates a layered risk-mitigation architecture combining tool-based controls with organizational and regulatory processes. The mapping protocol achieved an F1 score of 75.5% after majority voting. Overall, the study provides a practical mapping between enterprise AI risk categories and open-source mitigation capabilities, identifies where human oversight remains necessary, and presents a taxonomy-driven framework applicable to open-source and proprietary solutions.

Giving an AI agent the ability to send emails, query databases, or execute commands is useful--until the agent is tricked into doing something it shouldn't. Prompt injection, hallucinated reasoning, and unsafe tool calls form the primary attack surface for autonomous LLM agents. Existing defenses rely on software checks like system prompts or policy filters running on the same machine the attacker targets, offering no verifiable proof of execution. We introduce Niyam-AI, a framework that makes safety enforcement provable. At session start, permitted tools and constraints are locked into an Intent Contract committed via SHA-256. Every tool call is intercepted and validated by an isolated Judge model; upon passing, a zk-SNARK proof is generated via EZKL. The tool executes only after proof verification, allowing third parties to confirm enforcement without accessing Judge model weights. Evaluating Niyam-AI on 2,000 real-world scenarios from Agent-SafetyBench against NeMo Guardrails, Meta's Llama Prompt Guard 2, and OpenAI's GPT-OSS-Safeguard using 5-fold stratified cross-validation yields an F1 score of 88.5% with a 1.1% false-positive rate (bootstrap 95% CI: [85.19%, 91.88%], N=1000). McNemar's exact paired test confirms significant improvement: Niyam-AI wins 390 discordant scenarios against NeMo (vs 20 losses), 115 against Prompt Guard 2 (vs 13), and 384 against GPT-OSS-Safeguard (vs 19) with p < 0.0001 in all cases. Proof generation adds 2260.6 +/- 218.4 ms per approved action, while verification takes 53.1 +/- 11.8 ms. Niyam-AI provides a guardrail that is both highly accurate and mathematically verifiable--though this reflects a classifier adapted to Agent-SafetyBench evaluated against zero-shot baselines, a distinction discussed in Section IV.C.

Large language model (LLM) benchmark evaluations are routinely used to support claims about model safety, reliability, and deployment readiness. Yet most evaluations rely on a single access modality (model APIs), perform a single run per prompt, and report accuracy as the primary outcome metric, without accounting for conditions such as web search that may have effects on model behavior in deployment. We audit these assumptions for one of the most widely-used LLMs, comparing two modalities, ChatGPT's chat UI and OpenAI's API, with and without web search enabled. We use a stratified total sample of 401 prompts from two popular benchmarks, BBQ and SafetyBench, collecting 4,812 total responses across three repeated runs per prompt. Beyond standard performance measures, we evaluate model output dimensions including response consistency, response text similarity, citation grounding, and abstention behavior. For instance, chat UI responses were less accurate than API responses on both benchmarks with search disabled. Enabling web search reduced accuracy by up to 8 percentage points, and even reversed the direction of modality performance trends for one benchmark. Repeated runs of the same prompt produced inconsistent responses in up to 21% of prompts. The two modalities also grounded answers in different citations, and abstention behavior was also inconsistent across both modalities. These results illustrate that, even within a model family, reporting only simple accuracy metrics can obscure important forms of model behavioral variation relevant to AI safety assessments. We argue that AI safety evaluations should systematically account for modality, multi-run consistency, search conditions, and response-level behaviors to better reflect how deployed AI systems behave in practice.

Existing guidance identifies excessive agency, excessive permission, weak task-bound authorization, and inadequate agent controls as important risks. Control frameworks also describe capabilities for constraining, authorizing, observing, validating, and responding to agent activity. Yet security programs still need a way to manage persistent deployed instances that span components and outlive any one event. We propose the agentic posture vulnerability (APV) as a task-conditioned vulnerability-management abstraction: a durable record for a composed agent-control exposure. One posture may produce different runtime manifestations across tasks; APV links those manifestations to the invariant posture and remains open until authority is narrowed, a missing control is added, risk is accepted, or closure is verified. APV is not proposed as a new root-cause class of risk; it operationalizes existing excessive-agency, authorization, and control-composition weaknesses. We distinguish APVs from CVE-addressable product defects, OWASP Excessive Agency, Agent Baseline control outcomes, and the runtime authorization-execution gap. We then provide a field vignette, a thresholded definition, six recurring APV patterns, a vulnerability lifecycle, a minimum record, a control-and-closure matrix, tooling implications, and a testable research agenda.

Safety risks of AI are becoming increasingly evident in human interactions with AI technologies. The prominent approaches to evaluating these risks favor technical methods, such as model benchmarks and LLM simulations, often sidelining empirical research with human subjects. To examine this apparent gap in the acceptance of human research, we conduct an expert survey (n=93) and expert interviews (n=17) with AI Safety & Ethics (AISE) researchers from Technical, Sociotechnical, Governance, and Normative backgrounds. Our findings suggest that although there is a consensus that human research is valuable for generating evidence for AISE, its adoption and acceptance are constrained by perceived validity issues, tangible resource barriers, epistemic and personal preferences in methods, and infrastructural constraints from the broader research community. In particular, Technical researchers tend to value human research less and collaborate across disciplines less, suggesting an epistemic tension towards human methods. We propose recommendations for establishing the epistemic fit of human research within AISE and bridging the prohibitive limitations that researchers face, while avoiding performative 'human-washing'.

Language models differ in how safely they behave and these differences are measured by safety benchmarks. But aggregated benchmark scores are hard to trust and interpret, because benchmarks duplicate one another, correlate heavily, and models may sandbag when they detect evaluation. To address these issues, we draw on Item Response Theory (IRT), a statistical toolkit for measuring these latents from performance on items with inferred psychometric properties. We fit IRT models to eight safety benchmarks across 192 language models, the largest psychometric analysis of LLM safety evaluations to date, and contribute three results. First, we find that three interpretable factors of refusal strictness, truthfulness, and contextual harm explain most of the variance between models across benchmarks. Second, psychometrically selected items recover full benchmark scores with lower error than random subsets of the same size, and roughly ten adaptively chosen items suffice for several individual benchmarks, cutting evaluation cost by 97-99%. Third, IRT supports audits of individual models, showing that it can be used to detect naive sandbagging and changes of model behind APIs. Overall, we show IRT is a ready-made toolkit for reading, reducing, and auditing safety benchmarks, which we recommend frontier labs and evaluators adopt.

As AI systems become increasingly integrated into diverse interfaces and applications, model-centric audits are insufficient to address risks arising from interactions among system components and deployment environments. System integration has long been central to software audits in safety-critical domains such as aerospace. However, its role in AI auditing remains underexplored. Scanning through 4,259 documents, we present a scoping review of AI audits that treat system integration as a core tenet of evaluation (n = 58). Using reflexive thematic analysis, we analyze their elements, actors, enablers, and constraints. We find that the corpus represents an emerging yet still fragmented form of AI auditing: few existing measures target integration-specific risks; large gaps remain in meeting traditional audit expectations; and access to necessary information and resources significantly influences audit design. Nonetheless, integration can be categorized across three sites (inter-component, system-environment, and multi-system), each serving the functions of risk exploration, risk determination, coordination, and procedural regularity. Deviating from other types of evaluations, these audits assess qualities specific to system integration, including compatibility, completeness, and oversight. This review calls on the AI community to prioritize system integration as a core strategy for addressing AI risk, and to develop audit practices capable of capturing failures across components, environments, and systems beyond the reach of component-level evaluation.

Role-Playing Language Agents (RPLAs) are increasingly deployed in high-stakes applications such as healthcare assistance, customer support, and education, where maintaining consistent personas, ethical constraints, and behavioral coherence under adversarial pressure is critical. Existing evaluation approaches rely on static benchmarks or isolated single-turn prompts that fail to capture cumulative behavioral failures emerging over extended interactions. We present a modular multi-agent platform for adversarially stress-testing RPLAs through structured, multi-turn dialogue. The system coordinates three agents: a strategy-driven Interrogator Agent that applies six progressive adversarial strategies, a Target Agent representing the RPLA under evaluation, and an automated Judging Agent that scores behavior across role fidelity, drift, ethical deviation, and consistency dimensions. Through experiments across three personas and three LLM families, we demonstrate that multi-strategy adversarial evaluation reveals failure modes invisible to single-strategy testing, reducing overall robustness scores by 0.17--0.20 points on average. Cross-model validation confirms consistent degradation patterns across Llama-3.3-70B, GPT-4o-mini, and Claude-3.5-Haiku, with Authority Challenge and Emotional Manipulation emerging as the most effective attack strategies. Automated judging achieves strong human alignment ($r = 0.82$, Fleiss' $κ= 0.71$). This work is released as an open-source platform to support AI safety and reproducible RPLA benchmarking. While the framework enables systematic discovery of failure modes, we acknowledge potential ethical risks associated with adversarial testing methodologies and emphasize responsible usage for improving AI safety.

The AI Security Leaderboard is an independent benchmark that ranks the safeguards of frontier AI models from least to most secure. It tests models against the FAR$.$AI Minimal Standard for Safeguards, which represents a minimum bar for security: meeting it does not guarantee a secure model, but failing to meet it guarantees a lack of state-of-the-art security. Version 1.0 covers severe misuse requests across chemical, biological, radiological, nuclear, and explosive (CBRNE) threats and offensive cybersecurity. In this report, we tested four leading models for universal jailbreaks in the context of this minimal standard, and found more than a hundredfold difference in security. Claude Fable 5 and GPT-5.6 Sol held against every attack we ran, with no universal jailbreak found; we estimate they would likely cost more than $14,200 to jailbreak, if it is possible with this methodology at all. Meanwhile, we found hundreds of universal jailbreaks for Grok 4.5 and Gemini 3.1 Pro; each broke for under $300, with universal jailbreaks in Grok's weakest domain, cybersecurity, accessible for as little as $24. The gap is fixable: every weakness we found belongs to a known class of attack that already has a defense deployed in production models. The leaderboard will be updated on a rolling basis as new models are released, and the evaluation methodology and Minimal Standard will be periodically revised to take into account the latest capabilities and the state-of-the-art in safeguards. The leaderboard is available at leaderboard.far.ai.

Chain-of-thought (CoT) monitoring is an increasingly important component of AI safety stacks but relies on the assumption that a model's reasoning trace is informative about its actions. This work studies the limits of CoT monitoring through the lens of model poisoning. We demonstrate that backdoors can be implanted into reasoning models to elicit an attacker-chosen behavior while their CoT traces appear entirely benign. We find that these CoT-Hidden backdoors can be induced through simple fine-tuning recipes across reasoning-model architectures and sizes. When direct poisoning is ineffective, we introduce a curriculum training approach that progressively teaches the model to produce an attacker-chosen output while concealing the behavior from its reasoning traces. These findings suggest that CoT monitoring may be better framed as a question about the consistency between a model's reasoning trace and its final response than as anomaly detection within a trace. We further examine the mechanisms that allow models to suppress evidence of the target behavior from their reasoning traces. Causal interventions locate a trigger-conditioned activation pathway that does not depend on the visible reasoning, and residual stream verbalizations provide an anomaly warning near answer generation, but do not identify the trigger, target, or backdoor mechanism.

Autonomous agents are increasingly used to execute consequential tasks in environments governed by operational constraints, organizational policies, regulatory requirements, and technical standards. Their safety is therefore determined not by the correctness of individual actions, but by whether their overall behavior remains consistent with the rules and invariants of the systems in which they operate. As large language model (LLM)-based agents become more autonomous and increasingly delegate tasks across organizational boundaries, securing them evolves from a single challenge into a broad and interconnected landscape spanning the entire agentic stack. At the single-agent level, untrusted inputs through prompts, memory, retrieved knowledge, and tool interfaces create attack surfaces. In multi-agent settings, delegation and communication introduce challenges related to identity, trust, capability control, and decision transparency, while the underlying model routing and execution control plane remains vulnerable to manipulation and to unverified model provenance. Perhaps the most fundamental challenge is behavioral containment: sequences of individually permissible actions may collectively violate system-level constraints and safety invariants. At the broader level, supply-chain integrity, provenance, accountability, and end-to-end observability remain largely open problems. A common principle unifies these directions: security must become a verifiable property of the architectures, protocols, and runtimes that govern agent behavior, rather than an optional layer of guidance. Charting these challenges provides a roadmap toward trustworthy autonomous agent deployment.

As more responsibility is placed upon AI systems, it becomes increasingly important to guarantee that these systems are aligned with humanity. A common fear in AI safety is that human value is fragile -- that is, optimizing too heavily for an imperfect proxy to human values will lead to a catastrophic outcome. In this paper, we present a model of the alignment problem where an agent undergoes idealized alignment training that guarantees its value function satisfies a proxy condition before optimizing the world. Our primary results identify conditions on the human value function and the accuracy of several proxy conditions under which an agent with an $η$-catastrophic value function, one that is guaranteed to take the expectation of human value below $η$ in the limit of optimizing power, would be deployed. Our results highlight the danger of overoptimization and motivate AI designs that limit optimization pressure, such as quantilizers, rather than relying solely on pre-deployment training.

World models give embodied AI a predictive core: they compress observations into states, simulate action-conditioned futures, and enable planning beyond reactive control. This predictive layer, however, opens a new security boundary-compromise can propagate from data, sensors, prompts, or feedback into physical action. Rather than treating world models as an isolated component, this survey traces threats across their entire lifecycle-from data construction and representation learning, through state grounding and imagination, to trajectory evaluation, execution, and long-term adaptation via memory and tools. We show that familiar attack families: poisoning, backdoors, adversarial examples, sensor spoofing, prompt injection, trajectory manipulation, and supply-chain attacks take on distinct meanings when they corrupt world states, learned dynamics, affordance estimates, or safety costs. We also highlight a duality: world models can serve as runtime safety shields, yet when compromised or over-trusted they generate predictive safety illusions. The survey offers a lifecycle taxonomy, maps existing attacks to world-model security properties, outlines evaluation protocols for safety failures, and structures defenses across provenance, robust grounding, uncertainty-aware prediction, trajectory gating, feedback auditing, and deployment assurance.

Large language models have improved rapidly on tasks with verifiable answers, such as mathematics and programming. Much less is known about their ability to reason about what we call conceptual questions: questions for which no ground truth is realistically accessible and no widely accepted resolution methodology exists, but on which progress can still be made by debating arguments. Most philosophical questions are of this kind, as are central components of questions in AI safety, decision theory, and social choice. Our approach is based on the view that while bottom-line conclusions on such questions are hard to evaluate, individual contextualized arguments can be evaluated far more reliably. We therefore introduce a dataset of 951 argumentative critiques of 442 position texts, spanning topics from AI safety and decision theory to ethics and politics, with 1,458 ratings by six expert raters along dimensions including centrality, strength, correctness, and clarity. We propose two scoring functions and benchmark a range of models. Performance tracks general capability rankings.

The responsible development and deployment of artificial intelligence (AI) systems requires rigorous documentation of their constituent artifacts, e.g., datasets, model weights, training pipelines, and runtime dependencies. Although the Software Package Data Exchange (SPDX) 3.0 standard introduced native support for AI and dataset profiles, practical tooling capable of generating standards-compliant AI Bills of Materials (AIBoMs) in an automated and extensible manner remains scarce. This paper presents AIGen, a modular AIBoM generator that produces machine-readable, interoperable inventories of AI system components that comply with the SPDX 3.0 AI profile. AIGen works on top of the MLflow MLOps framework and combines mining heuristics with Large Language Models to generate AIBoMs. A plugin interface allows practitioners to extend the tool with domain-specific collectors without modifying the core codebase, supporting heterogeneous AI frameworks such as Hugging Face, PyTorch, and TensorFlow. AIGen is designed to facilitate compliance with the European Union AI Act, the NIST AI Risk Management Framework, and ISO/IEC 42001, providing a concrete, reusable foundation for transparent, accountable AI supply chain governance. Tool URL: https://github.com/danielebifolco/AIGen Tool Video: https://youtu.be/\_nAbXDWfVL4

Technological races create tension between speed and safety: actors may gain by moving faster than competitors, even when risky development is harmful. This is prominent in debates about artificial intelligence (AI), where competitive pressure is often argued to incentivise riskier, less safety-conscious development. We study this using a framed behavioural experiment based on an idealised AI race, in which paired participants repeatedly chose between Safe and Unsafe development under an uncertain time horizon. Unsafe development gave faster progress and higher immediate payoffs but accumulated private risk up to a treatment-specific maximum of 10%, 60%, or 90%; the race's competitive structure was held constant, and only this maximum risk varied. Neither the pre-registered comparison between risk levels nor the role of elicited risk preferences was supported by the data. Instead, exploratory analyses motivated by the task's repeated structure show that Unsafe behaviour is shaped less by risk preferences than by the evolving strategic state of the race: participants are more likely to choose Unsafe after their opponent does so, being ahead reduces Unsafe play while falling behind increases it, and first-round choices predict later behaviour. To interpret these effects we introduce a reduced evolutionary model with four strategies -- Always Safe, Always Unsafe, Conditionally Safe, and Conditionally Antisocial Safe -- which reproduces the treatment effect and shows how conditional Unsafe behaviour can be favoured by competitive race dynamics. Together, the experiment and model show that unsafe development can emerge from early behavioural momentum, opponent behaviour, and fear of falling behind, rather than from risk preferences alone, suggesting policy should focus on reducing competitive pressure and promoting cooperation in AI development rather than only individual risk.

Public services face growing pressure to adopt artificial intelligence (AI) to close the gap between rising demand and falling resources. That pressure has intensified with general-purpose AI (GPAI): AI built on large language models that can be directed by prompt alone to perform an effectively unbounded range of tasks. We argue that the properties that make these models attractive - their generality, accessibility, and low deployment cost - undermine the conditions under which AI safety has historically been pursued. The safety concepts that public service governance frameworks foreground - accuracy, bias, explainability, and accountability - were made tractable by narrow, purpose-built AI, and the mitigations that guidance documents prescribe presuppose exactly what GPAI removes. Accuracy cannot be quantified over unbounded outputs. Bias cannot be disaggregated when outputs are free-text judgements rather than categorical predictions. Explainability gives way to the appearance of explanation, and accountability erodes as outputs are optimized to persuade. We develop this through the case of policing, where the consequences of governance failure are most severe, and show why the same failure is likely to recur across other public services. The two mitigations that dominate policing AI strategy - expert evaluation and human-in-the-loop oversight - both rest on assumptions that GPAI violates. Safety assurance thus shifts from an intrinsic feature of building an AI tool to an optional add-on. We recommend a clear taxonomic distinction between narrow and general-purpose AI in governance documentation, a preference for technological parsimony, a pause on operational deployment of GPAI in policing until adequate evidence exists, and a coordinated national safety infrastructure with the authority to generate that evidence and determine when responsible deployment is achievable.

Cyber-capable AI agents combine language models with tools, memory, and execution environments to perform multi-step offensive-security tasks. Existing work separately measures cyber capability and catalogs attacks against agent components, but provides less guidance on containing a capable agent within the environments used to evaluate it. This review synthesizes five vulnerability classes at that boundary: multi-step offensive chains, objectives that conflict with sandbox boundaries, supply-chain and credential exposure, persistent command-and-control, and the speed of automated action. We use two separate preliminary incident records: the reported July 2026 Hugging Face/OpenAI evaluation breach and Anthropic's subsequent three-incident evaluation review. A comparative evidence protocol distinguishes record-specific factual claims from the shared systems lesson: the evaluation environment is itself part of the security boundary. Across the taxonomy and records, we examine controls for containment, privilege separation, provenance, and responder access, including the dual-use problem that defensive artifacts may also enable misuse. The review identifies practical priorities for evaluating cyber capability together with the security of the environment in which that capability is exercised.

Mainstream AI research emphasises capability growth and tolerates low failure rates when average-case performance is high. AI safety and alignment research has a different mission: to ensure that catastrophic failures never occur, under sparse evidence, adversarial dynamics, and fat-tailed risk. We argue that the two domains differ along two analytically independent axes---{\it capability profile}, demonstrating the absence of hazardous behaviours rather than the presence of positive capabilities, and {\it risk profile}, bounding worst-case outcomes under fat-tailed uncertainty rather than optimising average-case performance---and that mainstream epistemic practices are inadequate on both. Building on a structured synthesis grounded in a preregistered bibliometric baseline, we identify five cross-cutting gap dimensions in current alignment research, including the near-absence of institutionalised independent verification. To address these gaps, we propose {\sc ECAISA}, an Epistemic Code for AI Safety and Alignment comprising eight principles, a three-level scoring rubric, a four-level disclosure ladder that reconciles transparency with information-hazard and commercial-confidentiality constraints, a tiered applicability scheme, an information-hazard adjudication procedure, and seven anti-gaming mechanisms. {\sc ECAISA} does not certify that any AI system is safe; it constrains how safety-relevant research claims are documented, checked, and relied upon, with auditability rather than certification as its governance target.

The convergence of artificial intelligence (AI), digital sensing, and ubiquitous computing has created an unprecedented opportunity to transform myopia prevention from a reactive, population-based model into a proactive, precision-driven one. Despite evidence that half the world's population will be myopic by 2050, conventional approaches---school-based vision screening (Phase 1.0) and evidence-based risk factor management (Phase 2.0)---have proven insufficient. We review the emergence of Myopia Prevention and Control 3.0, defined by AI integration across three interconnected domains forming a closed-loop pipeline: (1) AI-driven risk stratification predicting individual-level risk through machine learning on multimodal data; (2) AI-enabled proactive monitoring via wearables, smartphones, and school screening networks; and (3) AI-powered personalized intervention with closed-loop feedback. We critically evaluate evidence across each stage, discuss challenges in data quality, model validation, ethics, and equity, and outline future directions including multimodal foundation models, digital twins, and causal machine learning.

Artificial intelligence (AI) systems are increasingly deployed in high-stakes domains such as healthcare, autonomous driving, finance, and education. While these systems offer powerful data-driven and adaptive capabilities, their complexity, rapid evolution, and dependence on dynamic data pipelines introduce new forms of engineering liability collectively referred to as AI Technical Debts (AITDs). AITDs arise from root causes spanning data governance, model implementation, algorithm design, architectural decisions, operational processes, documentation practices, and testing adequacy. Unlike conventional technical debt, many AITDs are latent and propagate across tightly coupled AI pipelines, leading to maintenance challenges, reliability degradation, and heightened safety or security risks. Guided by the principles of AI Trust, Risk, and Security Management (AI TRiSM), this study reinterprets technical debt through the interconnected dimensions of trustworthiness, focusing on AI safety and security technical debts. We conduct a systematic review of 60 primary studies and identify 31 distinct types of AITD, which are organized into a root-cause-oriented taxonomy comprising seven classes. The analysis examines how these debts map to 18 trust-related concerns, including 6 safety hazards and 12 security vulnerabilities. To support mitigation, the review synthesizes 34 actionable guidelines (8 safety and 26 security) targeting the prevention, detection, and reduction of AITDs across the AI lifecycle. Building on these findings, we introduce AITD-MAP, an integrated framework that connects the AITD taxonomy, quality and risk impacts, and mitigation strategies into a unified structure for risk-aware AI engineering. The framework aims to assist AI software engineers in making AI safety and security technical debts visible, understanding their root causes, and mitigating their presence.

High-impact generative AI makes catastrophic misuse a lifecycle-control problem, not merely a prompt-filtering problem. SAGE is a safety-first, authorization-separated architecture in which credible catastrophic-enablement risk constrains admissibility before utility, latency, or commercial objectives are considered. It combines signed release manifests, diverse detectors, robust risk envelopes, least-risk defaults, output checking, three-valued monitoring, protected audit chains, containment, and rollback. Formal results establish safety priority, conservative detector bounds, monotone release gating, tamper-evident records, and an authorization cut; two PRISM abstractions verify authorization separation and lifecycle invariants under explicit assumptions. A frozen, vendor-symmetric study sent 84 cases to each of four GPT, four Claude, and two Gemini snapshots: 840 calls yielded 794 target responses, 46 provider errors, and 449 successful judgments covering 375 responses. Eight snapshots had complete judged domain coverage. Harmful-compliance estimates were low; variation arose mainly from benign utility and safe redirection. Seven multiplicity-adjusted contrasts involving Claude, Gemini, or GPT-5 snapshots and the GPT-5 mini and GPT-5 nano snapshots were supported, while no tested contrast between the Claude or Gemini snapshots and GPT-5 or GPT-5.5 survived correction. The observed harmful-compliance range is a conservative, protocol-bound view from one generation per prompt with no tools, retrieval, history, or human adjudication; it is not an upper bound on operational assistance. A preregistered extension specifies how to test a wider best-worst gap using a locked split, repeated sampling, multi-turn and sandboxed-tool conditions, and domain-expert scoring.

A key question for AI safety is whether a language model expresses all of its reasoning in its output tokens. We demonstrate a concrete failure mode where frontier models exhibit invisible reasoning by leveraging semantically irrelevant filler tokens to improve performance on synthetic reasoning tasks. We evaluate 13 frontier language models across three tasks and find that many models benefit significantly from filler tokens, with accuracy improvements of up to 13 percentage points. The benefit depends on which tokens are used and differs across models. We further show that filler tokens enable Claude Opus 4.5 to satisfy a hidden modular arithmetic constraint without sacrificing accuracy on its primary task, demonstrating that invisible reasoning can serve objectives entirely invisible to CoT monitoring. Reinforcement learning gives Qwen3-235B strong preferences over filler token content, but neither RL nor supervised fine-tuning produces a filler token benefit that persists at test time. Our results indicate that frontier models already perform consequential computation with no interpretable trace in their output tokens.

Agent security is widely treated as a question about action content. Defenses ask whether an instruction looks malicious. Benchmarks ask whether an agent performs a harmful sounding action. \textbf{We argue that agent security is fundamentally a contextual problem, and that the current content based framing systematically misdefines it.} A command to ``delete user data'' might be a routine administrative request or a prompt injection attacking production systems, and the content alone cannot distinguish the two. Authorization context can. Across every injection task in AgentDojo and WASP, the same action is one an authenticated user would plausibly request in a routine workflow, which makes the conflation a structural property of evaluating security through content. We operationalize contextual security through four properties that must hold jointly and be evaluated continuously across the agent's trajectory. Source Authorization asks who issued the command. Task Alignment specifies the agent's authorized objective. Action Alignment evaluates whether each action serves that objective. Data Isolation governs information flows across privilege boundaries. Under this reframing, indirect prompt injection becomes a Source Authorization violation. Snapshot benchmarks are structurally incapable of evaluating Data Isolation. Existing defenses are reorganized around the property they actually approximate. The contextual reframing changes which defenses are coherent, which evaluations measure something useful, and which attack patterns evaluation can see at all.

LLM-driven autonomous agents are reshaping offensive security. Unlike traditional penetration-testing tooling - deterministic, narrowly scoped, and operated by trained practitioners - agentic security tools exhibit indeterminacy along three independent dimensions. First, their actions are drawn from a non-deterministic policy whose outputs resist both ex-ante and ex-post explanation. This complicates incident attribution and pre-deployment safety reviews. Second, their impact is open-ended due to their non-deterministic actions, agency of utilized models, and opaque LLM supply-chains. Third, their user population is indeterminate in both size and required skill: the operating skill floor for using or developing offensive capabilities has dropped sharply. These three properties are linked thematically, but are not derivable from one another. Combined with the structural cost asymmetry between offense and defense, they enable the industrialization of offensive capability. The net short-term effect favors attackers, even if the same technology may, in the long run, democratize access to defensive practice. Existing dual-use cybersecurity and AI-ethics frameworks struggle to address this combination. Our work analyzes how moral attribution becomes diffuse between users, tool-makers, and third parties when employing autonomous AI agents for offensive security. We also examine the stakeholder impact of this technology and provide stratified recommendations.

As AI agents begin to automate AI R&D, we need ways to assess whether their outputs are safe to deploy, even when the agents themselves may be untrusted. AI control offers one such approach: rather than trusting the agent, it treats it as a potential adversary and uses a monitor to detect covert sabotage before deployment. We evaluate AI control for automated AI R&D with ResearchArena, a framework spanning four long-horizon tasks: safety post-training, capabilities post-training, CUDA-kernel optimization, and inference-server optimization. Because the deliverable in AI R&D is an artifact that will be deployed and run, we pair each main task with two kinds of hidden side task: an embedded side task that sabotages the submitted model, adapter, kernel, or server itself, and an independent side task that takes a covert out-of-scope action in the sandbox while leaving the artifact honest. We evaluate frontier agents at both sabotage and monitoring, across four monitor types that vary along two axes: whether the monitor may execute and probe the produced artifact, and whether it sees the agent's chain-of-thought. We find that sabotage hidden in the training data is the hardest to catch, flagged fewer than half the time. Letting the monitor run experiments on the artifact, rather than only read the trajectory, helps, but it is not enough: monitors still miss embedded sabotage by inspecting only the surface, by noticing the anomaly but explaining it away, or by probing the artifact with the wrong test. We release ResearchArena as a modular framework for evaluating sabotage and control in automated AI R&D.

Current AI safety discourse still focuses disproportionately on visible failures, including obvious harms, dramatic misuse, and hypothetical catastrophic scenarios. That focus is incomplete. In deployed systems, many of the most consequential failures are quieter: plausible rather than spectacular, distributed across components rather than localized in a single output, and normalized by workflows before they are recognized as hazards. We argue that a central safety challenge in modern AI systems is increasingly not only whether a model emits a harmful response, but whether the broader socio-technical system preserves the conditions under which errors remain visible, contestable, containable, and recoverable. We propose a five-layer framework for diagnosing these hidden risks: (1) epistemic integrity, concerning whether evidence and uncertainty are represented honestly enough to support calibrated reliance; (2) control integrity, concerning whether authority, permissions, and action boundaries remain robust under attack and optimization; (3) temporal integrity, concerning whether safety holds across sessions, memory updates, and deployment drift; (4) organizational integrity, concerning whether institutions retain the capacity to audit, assign responsibility, and intervene effectively; and (5) ecosystem integrity, concerning whether AI systems preserve rather than erode the information environment on which future oversight depends. Across these layers, we identify under-recognized risk patterns, including overreliance, uncertainty and legitimacy laundering in retrieval, prompt injection, reward hacking, memory poisoning, evaluation deception, fictional human oversight, synthetic evidence pollution, and model collapse. We conclude with design and governance recommendations and a research agenda for shifting AI safety from model-centric evaluation toward socio-technical reliability.

The rapid growth of Artificial Intelligence-generated content (AIGC) is reshaping video production and circulation, exposing children to an increasing volume of AIGC videos. Unlike traditionally produced videos, AIGC videos often exhibit greater uncertainty in visual details, narrative coherence, and content expression, which may introduce developmentally inappropriate risks for children. However, existing video safety research is largely designed for general violation detection from an adult perspective and remains insufficient for identifying the fine-grained, implicit, and context-dependent risks that children may encounter when viewing AIGC videos. To address this gap, we study child-oriented AIGC video reviewing, making three contributions. First, we construct CAVSR, a benchmark of 605 real-world videos collected from multiple platforms, and develop a hierarchical risk taxonomy comprising 6 top-level categories and 26 fine-grained labels to support systematic evaluation of children's viewing risks. Second, we propose QVRS-E, a knowledge- and experience-augmented video reviewing framework that combines multi-agent collaboration with expert and experiential knowledge to support targeted evidence acquisition and fact-grounded reviewing decisions. Third, extensive experiments demonstrate that our method significantly enhances the reviewing of child-related risks integrated with vision-language models, and yields more robust review reports.

The transition from stateless generative models in artificial intelligence to stateful, autonomous agents represents an architectural evolution that, while providing the capabilities of long-term planning and the automation of enterprise workflows, also represents the introduction of a new form of security threat, the Chronos Vulnerability. The Chronos Vulnerability represents the threat of memory-based attacks, including the Memory Injection Attack (MINJA) and the sleeper agent, in which the internal belief system of the autonomous agent is compromised, effectively decoupling the attack vector from the final catastrophic event. This study formalizes the threat model for persistence-based attacks and the threat of Dynamics Blindness in the context of the World of Workflows benchmark, demonstrating that traditional endpoint content filters are insufficient for the current stateful architecture. Consequently, this study synthesizes a defense-in-depth landscape, categorizing emerging frameworks such as diagnostic trajectory guardrails (AgentDoG), formal temporal verification (Agent-C), immunological memory consensus (A-MemGuard), and hardware-anchored trust via GPU-based Trusted Execution Environments (TEEs) and Zero-Trust memory architectures.

Large language models (LLMs) serving as planners in tool-using autonomous agents introduce dynamic reliability risks in multi-turn execution. While single-turn safety mechanisms are relatively mature, extended interactions reveal structural vulnerabilities where initial alignment degrades over time. This paper empirically characterizes two observed failure modes across multiple state-of-the-art LLMs: Safety Drift, the gradual erosion of declared safety intent leading to constraint-violating actions (e.g., textual refusal followed by reconnaissance and unsafe execution), and Operational Hallucination, persistent repetitive tool calls indicative of flawed state perception (e.g., livelocks even in legitimate tasks). Through controlled multi-turn evaluation on high-stakes ethical dilemmas, malicious requests, and benign controls, we quantify these phenomena using declaration-action gap and livelock metrics, demonstrating their cross-model prevalence under direct execution protocols. Root-cause analysis attributes the instabilities to the decoupling of reasoning context from execution state in current agent loops. We propose an Action-Aware Supervision Layer - a lightweight, plug-and-play architectural blueprint incorporating intent-action consistency checks, runtime state tracking, and forced termination primitives. Post-hoc simulation on captured failure trajectories shows the layer can intercept observed violations without false positives on benign cases. This work advances agent reliability by shifting focus from linguistic safeguards to enforceable architectural mechanisms for responsible agentic AI.

The increasing adoption of Large Language Models (LLMs) as AI components in modern software systems introduces distinct security risks to the software supply chain. While many considerations and safety mechanisms are in place for components of the traditional software supply chain, the recent rapid adoption of AI components and platforms has overlooked these hard learned lessons. Selecting and integrating AI models without clear guidance on how these choices affect system security may leave applications vulnerable to threats, such as malicious components, data leakage, and unintended behavior. The goal of this study is to understand practitioners' decision making process and security considerations in selecting and integrating AI components through an exploratory semi-structured interview study. Toward this goal, we conducted semistructured interviews with 22 software developers, architects, and AI practitioners across diverse organizations about how they integrate AI components into their software. Our analysis finds that practitioners' model selection is predominantly driven by functional criteria, including performance, accuracy, cost, and specific features, e.g., tool calling or multimodal support, while security is rarely considered as an evaluation criterion. We observe a consistent lack of security concern throughout the AI component integration process, with established software supply chain lessons overlooked or ignored. The industry is repeating the historically costly mistakes of early software dependency management, prioritizing rapid reuse and availability over security and provenance. We distill our findings into actionable recommendations for AI adopters, model providers, and researchers, advocating for a proactive, security-by-design approach that integrates security evaluation into component selection and sustains it throughout the software development lifecycle.

Artificial intelligence (AI) systems are now ubiquitous across domains such as security, finance, healthcare, consumer technology, and large-scale cloud services, where they process massive volumes of data and make consequential decisions daily. This widespread adoption has created a broad attack surface through which adversaries can manipulate, evade, extract information from, or otherwise subvert deployed models. Depending on system design and exposure, attackers may have very different forms of access: some observe only final decisions, while others receive confidence scores, intermediate representations, or even full model parameters. While previous surveys typically organize evasion attacks into white-box, gray-box, and black-box categories based on the attacker's knowledge of model internals (architecture, parameters, gradients), this taxonomy often conflates different deployment scenarios that provide vastly different output signals, all labeled as ``black-box'' despite enabling fundamentally different attack strategies. Understanding how evasion attack strategies adapt to the specific information signals returned by deployed systems is critical for organizations making procurement and deployment decisions. To address this gap, we introduce the Signal-based Model Access Risk Taxonomy (SMART), a deployment-oriented framework that classifies attacker access according to the nature and richness of the information signals available from deployed AI systems. Using this taxonomy, we provide a structured overview of evasion attacks across progressively richer levels of information exposure, highlighting how deployment interfaces influence attack capabilities and informing more secure AI deployment and procurement decisions.

Frontier AI companies have published capability thresholds that differ substantially, making it difficult for third parties to verify whether a threshold has been crossed or to compare requirements across companies. Moreover, without common minimum thresholds, risk mitigation may be inconsistent, creating a potential race to the bottom in safety standards. We develop a methodology for deriving harmonized thresholds across three risk domains. For misuse risks (cyber and biological), we take expected harm as the key primitive and use an explicit risk-modeling approach that accounts for risk channels and model release conditions. For automated AI R&D, we base our proposed threshold on the observed rate of AI progress rather than expected harm. Our analysis expands upon prior work and highlights existing empirical gaps and limitations.

Most medical AI benchmarks measure whether a model knows the correct answer. MedFailBench asks a different question: which safety boundary failed? We present a clinician-built synthetic benchmark and failure atlas that labels medical AI errors by severity (1--5) and safety gate type (missed urgent escalation, unsafe remote dosing, unsafe discharge reassurance, evidence fabrication, unsafe protocol execution, source support gap). The current public release (v0.2.1) contains 44 clinician-reviewed synthetic cases with severity annotations, a live HuggingFace leaderboard preview, a safety gate taxonomy, a clinical severity rubric, and an automated pipeline for archiving model-response screening runs. No patient data, clinical validation claims, or model rankings are included. MedFailBench is released under Apache-2.0 and CC-BY-4.0 and carries the Zenodo DOI 10.5281/zenodo.21205535.

AI software development agents are increasingly capable of modifying infrastructure and security critical systems, creating risks where an agent completes its assigned task while covertly weakening safeguards through actions such as broadening permissions, degrading logging, or introducing persistence mechanisms. While frontier laboratories may deploy sophisticated monitoring pipelines, many organizations and individual users adopting coding agents lack the resources and governance maturity required to maintain complex learned monitor ensembles. We study this problem as infrastructure level sabotage under task success in the infrastructure-as-code setting from ControlArena and compare trained and untrained monitors across synchronous and asynchronous deployment modes. We introduce an Information Flow Graph (IFG) monitor that analyzes structural security regressions using control-flow and data-flow graph diffs alongside raw code diffs. In asynchronous evaluation, an untrained git diff monitor misses 11.6% attacks at 1% false positive rate auditing budget, our untrained IFG monitor reduces this to 3.5%, and the trained Async RF monitor achieves below 1%. Unlike the trained Async RF monitor, IFG can also operate synchronously as a pre-deployment safeguard that blocks suspicious infrastructure changes before execution or merge. In synchronous mode, IFG realtime rollback reduces joint main task and covert task success from 74.4% to 0.0% with no measurable reduction in legitimate task completion. These results suggest that untrained structural monitors provide a practical and auditable path toward democratizing deployment safety for organizations seeking trustworthy adoption of advanced AI agents.

As automated decision-making and data-driven technologies pervade society and are used to manage consequential outcomes, understanding the technology's capabilities, limitations, and attendant risks in context requires analysis of full sociotechnical systems. Sociotechnical analysis of risks in highly complex systems provides clear lessons for the design and evaluation of AI systems, transcending a technical focus on reliable or "responsibly designed" components to understand risks at a systems level. Human-made catastrophes have been studied for decades because of the severity of these events: consider Chernobyl, Three Mile Island, Fukushima-Daiichi, Bhopal, the Challenger disaster. A common misconception is that these kinds of events are freak accidents, resulting from the inherently unforeseeable interactions in complex systems. Closer examination reveals that the risks and hazards were well-known beforehand but not acted upon due to social structural, political and economic factors. We outline several areas where the development and use of AI can benefit from learning these unlearned lessons: improved risk perception, communication, and analysis at the organizational level; traceability of requirements and responsibilities; and holistic approaches to responsibility and safety that include social and organizational dynamics as first-order engineering concerns. For each area, we offer concrete unlearned lessons and exemplify how they led to failure in prior accidents as well as examples of how these lessons remain unlearned for modern computing systems, particularly AI.

Artificial Intelligence (AI), especially Generative AI (GenAI), adoption has increased in industries significantly in recent years. However, the use of these models may also expose systems to new forms of cyberattacks by different malicious actors -- adversarial prompt attack (APA) being one of the most prominent examples of such threats. This paper presents the implementation of an Adversarial Prompting Framework (APF) for a comprehensive assessment of AI safety. The framework systematically evaluates the resilience of the AI model through the generation of structured adversarial prompts at multiple sophistication levels, from direct harmful requests to advanced encoding-based attacks. Our implementation demonstrates the practical application of this methodology in enterprise environments, providing automated testing capabilities with quantitative security assessment metrics. The results indicate significant variations in the model vulnerabilities across different attack vectors, with encoded prompts presenting the highest success rates in bypassing safety mechanisms.