Safety papers

Browse recent papers, open the full write-up, or filter the archive by topic.

287 papers found

ReviewPaperPublishedTopics

6.8Can Agents Deceive? Evaluating Reasoning and Deception in ParliamentBench using a Social Deduction GameUnderstanding LLMs' capabilities in deception and reasoning is crucial for ensuring their safe deployment in high-stakes environments.Jul 31, 2026LLM Agent Reasoning

6.8Adversarial Pragmatics for AI Safety Evaluation: A Diagnostic Framework and Seed Benchmark for Language-Mediated ControlThis work addresses critical gaps in the evaluation of language models, particularly in understanding the nuances of model behavior under ambiguous instructions, which is essential for improving AI safety.Jul 31, 2026LLM Evaluation Benchmark

7.5Towards Trustworthy Embodied Intelligence: A Systems Framework and Graded Trustworthiness LevelsAs embodied AI systems become more prevalent, establishing a clear framework for trustworthiness is crucial for ensuring safety and reliability in real-world applications.Jul 31, 2026Robotics Safety Other

6.8State-Dependent Safety Failures in Multi-Turn Language Model InteractionUnderstanding and diagnosing safety failures in multi-turn interactions is crucial for deploying language models in real-world applications where sustained interaction is common.Jul 31, 2026LLM Safety Evaluation

6.8The Alignment Target Problem: Divergent Moral Judgments of Humans, AI Systems, and Their DesignersUnderstanding how moral judgments differ between humans, AI systems, and their designers is crucial for developing AI systems that align with human values, especially in high-stakes domains.Jul 31, 2026Alignment Evaluation

6.8Why Public Service AI Governance Frameworks Risk Failing in the Age of General-Purpose AI: Lessons from PolicingUnderstanding the limitations of current AI governance frameworks is crucial for AI engineers and researchers to ensure safe and responsible deployment of general-purpose AI in public services.Jul 30, 2026Safety Survey Other

7.2Inspect India Evals: An Open Benchmarking Framework for Evaluating Large Language Models in the Indian Linguistic and Cultural ContextThis framework is crucial for developing and assessing LLMs that are fair, safe, and accurate in the diverse Indian context, which is often overlooked by existing benchmarks.Jul 29, 2026LLM Evaluation Benchmark

6.8AIR-BENCH Live: An Evolving Safety Benchmark for Foundation ModelsAs AI models and regulations rapidly evolve, a dynamic benchmark like AIR-BENCH Live helps ensure ongoing safety and compliance, which is crucial for developers and policymakers.Jul 28, 2026Benchmark Safety Evaluation

6.8EvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety FailuresUnderstanding and improving the evaluation and safety of LLMs is crucial for developing reliable AI systems, making this framework potentially valuable for researchers and practitioners focused on AI safety and evaluation.Jul 28, 2026LLM Alignment Evaluation

6.8Separating Capability from Permission: A Governance Framework for Agentic AI Autonomy LevelsUnderstanding and managing the distinction between what AI systems can do and what they are allowed to do is crucial for safe and effective deployment in real-world applications.Jul 28, 2026Agent Safety Other

7.2AI-generated Images Challenge Visual Trust in High-risk ScenariosUnderstanding the limitations of current detection systems in high-risk scenarios is crucial for developing more reliable AI tools to maintain visual trust and ensure public safety.Jul 28, 2026Vision Language Safety Benchmark

6.8Epistemic Norms for AI Safety and Alignment ResearchAI engineers and researchers should care because ensuring the safety and alignment of AI systems is crucial to prevent catastrophic failures, and this paper proposes a framework to improve the reliability of safety-related research.Jul 28, 2026Alignment Safety

7.2Risk Governance for Generative AI Mental Health Support: A Multi-Turn Safety ArchitectureThis work is crucial for deploying LLMs in sensitive areas like mental health, ensuring safer and more effective interactions.Jul 28, 2026LLM Safety Architecture

6.8Opaque Epistemic Mediation: How LLM Deployment Configurations Shape the Validation of Pseudo-ScienceUnderstanding how deployment settings affect LLM outputs is crucial for ensuring the reliability and accountability of AI systems used in sensitive contexts.Jul 27, 2026LLM Evaluation Safety

6.5Benchmarking Large Language Models on Multi-Sensor Physical Hazard AssessmentUnderstanding the limitations of language models in multi-sensor environments is crucial for their deployment in safety-critical systems.Jul 24, 2026LLM Evaluation Safety

7.0OpenSkillRisk: Benchmarking Agent Safety When Using Real-World Risky Third-Party SkillsUnderstanding and improving agent safety when using third-party skills is crucial for deploying reliable AI systems in real-world applications.Jul 24, 2026LLM Agent Safety

7.2SafeHarbor: Defining Precise Decision Boundaries via Hierarchical Memory-Augmented Guardrail for LLM Agent SafetyAs LLMs evolve into autonomous agents, ensuring their safe operation in real-world environments is crucial to prevent misuse and maintain user trust.Jul 24, 2026LLM Agent Security

6.5Safeguards for Speech2Speech LLM-Assistants: A Case Study in Automotive ApplicationsUnderstanding the limitations of S2S guardrails is crucial for developing reliable and safe conversational AI systems in vehicles.Jul 24, 2026Safety Agent LLM

6.8Playing Along: Learning a Double-Agent Defender for Belief Steering via Theory of MindUnderstanding and improving AI's ability to model and manipulate beliefs is crucial for developing robust systems capable of safe interactions in adversarial settings.Jul 24, 2026LLM Agent Reasoning

6.5The Ethics of Autonomous AI Agents for Offensive SecurityUnderstanding the ethical implications of autonomous AI in security is crucial for developing responsible AI technologies and policies.Jul 23, 2026LLM Agent Security

6.8Will the Agent Recuse, and Will It Stop? Measuring LLM-Agent Compliance with In-Band Governance Signals at the Access Door and Mid-FlightUnderstanding and improving LLM agent compliance with governance signals is crucial for safe and controlled deployment in real-world applications.Jul 23, 2026LLM Agent Security

7.2JailMeter: An Evidence-Based Evaluation Framework for Jailbreak Attacks on Large Language ModelsUnderstanding and evaluating jailbreak attacks is crucial for improving the safety and alignment of large language models, which are increasingly used in sensitive applications.Jul 23, 2026LLM Evaluation Security

7.2NEXUS: Structured Runtime Safety for Tool-Using LLM AgentsAs LLM agents increasingly perform high-impact actions, ensuring their safe operation is crucial for preventing unintended consequences in real-world applications.Jul 23, 2026LLM Agent Safety

6.8JANUS: Foreseeing Latent Risk for Long-Horizon Agent SafetyImproving agent safety by predicting and preventing risks can significantly enhance the reliability and trustworthiness of AI systems in real-world applications.Jul 23, 2026Agent Safety

7.5Matching Ranks Over Probability Yields Truly Deep Safety AlignmentAs open-source LLMs become more prevalent, enhancing their safety against misuse is crucial for responsible AI deployment.Jul 23, 2026Alignment Safety LLM

7.5SENTINEL: A Multi-Level Formal Framework for Safety Evaluation of Foundation Model-based Embodied AgentsAs AI systems increasingly operate in physical environments, ensuring their safety through formal methods is crucial for real-world applications and public trust.Jul 22, 2026Safety Evaluation Agent

6.8The safety failures we are not instrumenting: a perspective on hidden safety-critical challenges in modern AI systemsUnderstanding and addressing hidden safety challenges is crucial for developing reliable AI systems that can be trusted in real-world applications.Jul 22, 2026Safety Alignment Survey

6.8Engineering Trustworthy Agentic AI for Critical SystemsUnderstanding and ensuring the trustworthiness of agentic AI is crucial for their safe deployment in critical systems where failures can have significant consequences.Jul 22, 2026Agent Safety Security

7.2Operational Hallucination and Safety Drift in AI AgentsUnderstanding and mitigating reliability risks in AI agents is crucial for safe deployment in real-world applications, especially in high-stakes scenarios.Jul 22, 2026LLM Agent Safety

6.8Evaluating medical AI under missing information: same-provider judges and human raters change apparent safetyUnderstanding how evaluator biases affect AI safety assessments is crucial for developing reliable medical AI systems.Jul 22, 2026LLM Evaluation Safety

6.8Agents in the Wild: Where Research Meets DeploymentUnderstanding the deployment challenges and solutions for agentic systems is crucial for AI engineers and researchers aiming to implement these systems in real-world applications.Jul 22, 2026LLM Agent Reasoning

7.5The Economics of Autonomy: Real-Time Risk Indexing for Insurable AI-Driven 6G SystemsAs 6G networks evolve, effective risk management frameworks like GIRAF are crucial for ensuring safety and reliability in increasingly autonomous systems.Jul 22, 2026Security Safety Infra

7.5Estimating Rare Events in Language Models with Proper EvaluationUnderstanding and estimating rare failures in language models is crucial for ensuring their reliability, especially in high-stakes applications, making this work relevant for both researchers and practitioners in AI safety.Jul 22, 2026LLM Evaluation Safety

6.8Specifying the Delegated-Autonomy Boundary: Requirements Engineering for Agentic AIUnderstanding and specifying the delegated-autonomy boundary is crucial for developing safe and effective AI systems that operate with varying levels of autonomy.Jul 21, 2026Agent Safety

7.8RAIL Guard: Closing the Evaluation-to-Remediation Gap in Responsible AI for LLM AgentsThis research addresses critical gaps in responsible AI by providing a systematic method to enhance the safety and utility of LLM outputs, which is essential for organizations deploying these technologies.Jul 21, 2026LLM Agent Evaluation

7.2A Dual-Hypothesis Reasoning Framework for LLM GuardrailsThis framework could improve the safety and interpretability of LLMs, which is crucial for their deployment in sensitive applications.Jul 21, 2026LLM Reasoning Safety

6.8Decoupled Alignment for Robust Plug-and-Play AdaptationThis approach could simplify the process of making language models safer for deployment by avoiding the need for extensive retraining.Jul 20, 2026LLM Alignment Safety

7.0Human-Inspired Neuro-Symbolic World Modeling and Logic Reasoning for Interpretable Safe UAV Landing Site AssessmentThis work is relevant for AI engineers and researchers interested in improving the safety and interpretability of UAV operations in unstructured environments, which is crucial for real-world applications.Jul 20, 2026Reasoning Interpretability Safety

6.8SeerGuard: A Safety Framework for Mobile GUI Agents via World Model PredictionAs mobile GUI agents become more prevalent, ensuring their safe operation is crucial to prevent potentially irreversible errors.Jul 20, 2026Agent Safety Security

7.5When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk SpaceUnderstanding the difference between text-based safety and real-world risks is crucial for the safe deployment of AI systems in physical environments.Jul 18, 2026LLM Safety Evaluation

6.8ToolAlignBench: Investigating Alignment Conflicts in Tool-Calling Enabled LLMsUnderstanding alignment conflicts in LLMs is crucial for deploying these models in sensitive environments, where conflicting values can lead to unpredictable and potentially harmful actions.Jul 18, 2026LLM Alignment Safety

6.5Unsafe at any AUC: Unlearned Lessons from Sociotechnical Disasters for Responsible AIAI engineers and researchers should care because understanding systemic risks and organizational dynamics can prevent future AI-related failures.Jul 18, 2026Safety Survey Other

6.8Governing Artificial Intelligence: Public Preferences and Regulatory OptionsUnderstanding public preferences can guide policymakers in aligning AI regulations with societal values, potentially leading to more effective and accepted governance frameworks.Jul 18, 2026Survey Safety Alignment

6.8Human-In-The-Loop Machine Learning for Safe and Ethical Autonomous Vehicles: Principles, Challenges, and OpportunitiesUnderstanding how to effectively incorporate human oversight in autonomous vehicle systems is crucial for improving safety, reliability, and ethical compliance in real-world applications.Jul 18, 2026Survey Safety Robotics

7.5Democratizing Agent Deployment Safety: A Structural Monitoring ApproachAs AI agents increasingly modify critical systems, ensuring their safe deployment is essential for organizations lacking sophisticated monitoring capabilities.Jul 18, 2026Security Safety Coding

7.5Towards an Intention Abstraction Layer for Autonomous Industrial SystemsThis work addresses a critical gap in ensuring that autonomous systems align with human intentions, potentially reducing operational failures in complex industrial environments.Jul 18, 2026Agent Safety Interpretability

7.2SingGuard-NSFA: Extensible Guardrails for Agentic AI via Generative Reasoning and Real-Time ClassificationAs AI systems become more autonomous, ensuring their security against operational threats is crucial for safe deployment in real-world applications.Jul 16, 2026Agent Security Safety

6.8Post-Deployment Accountability in AI Governance: A Cross-Regulatory Empirical Analysis of AI IncidentsUnderstanding and improving AI governance is crucial for ensuring accountability and compliance in high-stakes AI deployments, which is essential for both developers and regulators.Jul 16, 2026Safety Other

7.2The Joint Effect of Quantization and Sampling Temperature on LLM Safety Alignment: A Factorial AnalysisUnderstanding the joint effects of quantization and sampling temperature on LLM safety is crucial for deploying models that are both efficient and aligned with safety standards.Jul 16, 2026LLM Alignment Evaluation

6.8AI-Native Insurance for Agentic AI: Pricing, Underwriting, and End-to-End AutomationAs AI systems become more autonomous, understanding how to insure them effectively is crucial for managing risks and ensuring responsible deployment.Jul 16, 2026Agent Security Safety