CompanyAnthropic8 recent entries2 Sept 2026EvoFlint: An Evolutionary Atlas of Multi-Turn LLM VulnerabilitiesarXiv:2609.00487v1 Announce Type: cross Abstract: Frontier language models that refuse harmful single-turn prompts often comply when the same intent is reached gradually over many turns, making multi-→3 Sept 2026FUSE: An Evaluating Framework for Dangerous Capabilities of LLMsarXiv:2609.02168v1 Announce Type: new Abstract: Fragmented safety evaluation undermines the governance of dangerous AI capabilities. We present a modular framework that evaluates each model through th
CompanyOpenAI8 recent entries12 May 2026Less Diverse, Less Safe: The Indirect But Pervasive Risk of Test-Time Scaling in Large Language ModelsarXiv:2510.08592v3 Announce Type: replace-cross Abstract: Test-Time Scaling (TTS) improves LLM reasoning by exploring multiple candidate responses and then operating over this set to find the best out→19 May 2026Multilingual jailbreaking of LLMs using low-resource languagesarXiv:2605.18239v1 Announce Type: cross Abstract: Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails. We investigate whether multi-turn conversation→26 May 2026AI Content Moderation in Therapy ConversationsarXiv:2605.25454v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly being used for emotional support. They are also being developed for formal therapy purposes. However, LL→3 Jun 2026Cost-Aware Query Routing in RAG: Empirical Analysis of Retrieval Depth TradeoffsarXiv:2606.02581v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) faces a fundamental three-way tension: deeper retrieval improves factual grounding but inflates token costs and e→24 Jul 2026IssueTrojanBench: Benchmarking AI Coding Agents Against Malicious Issue RequestsarXiv:2607.20759v1 Announce Type: cross Abstract: AI coding agents powered by LLMs are increasingly integrated into real-world software development, where they generate, edit, and execute code with au→10 Aug 2026NiyamAI - An Intent-Bound AI Agent with Cryptographically Verifiable Guardrails using Zero-Knowledge ProofsarXiv:2608.07167v1 Announce Type: new Abstract: 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 should→11 Aug 2026When LLM Agents Negotiate: Private Information and Dynamic Bargaining in Supply ChainsarXiv:2608.07538v1 Announce Type: new Abstract: As LLM agents move from decision support to autonomous procurement, firms need to know whether delegated negotiators create value, divide it predictably→1 Sept 2026Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety PipelinesarXiv:2608.29589v1 Announce Type: new Abstract: Text-to-image (T2I) safety guardrails fail to generalize equitably to non-standard dialects. Evaluating 23,080 paired prompts across five English dialec
CompanyGoogle8 recent entries11 Aug 2026When Grammar Guides the Attack: Uncovering Control-Plane Vulnerabilities in LLMs with Structured OutputarXiv:2503.24191v4 Announce Type: replace-cross Abstract: Content Warning: This paper may contain unsafe or harmful content generated by LLMs that may be offensive to readers. Large Language Models (L→11 Aug 2026RouteGuard: Certifying Routing Gain in LLM Multi-Agent Systems When Complementarity Is Not EnougharXiv:2608.07583v1 Announce Type: cross Abstract: Multi-agent LLM systems route among model-backed advisors, yet a deployer rarely knows before shipping whether routing will help at all. Prevailing ro→14 Aug 2026Reasoning Jury: Multi-Model Consensus for Evaluating Reasoning TracesarXiv:2608.12585v1 Announce Type: new Abstract: Improving reasoning LLMs requires the ability to judge the quality of long reasoning traces for effective reasoning data curation, strong training signa→26 Aug 2026ADVERSA: Measuring Multi-Turn Guardrail Degradation and Judge Reliability in Large Language ModelsarXiv:2603.10068v2 Announce Type: replace-cross Abstract: Most adversarial evaluations of large language model (LLM) safety assess single prompts and report binary pass/fail outcomes, which fails to c→1 Sept 2026T-MAP: Red-Teaming LLM Agents with Trajectory-aware Evolutionary SearcharXiv:2603.22341v2 Announce Type: replace-cross Abstract: While prior red-teaming efforts have focused on eliciting harmful text outputs from large language models (LLMs), such approaches fail to capt→4 Sept 2026Measuring Harmfulness of Computer-Using AgentsarXiv:2508.00935v3 Announce Type: replace-cross Abstract: Computer-using agents (CUAs), which can autonomously control computers to perform multi-step actions, might pose significant safety risks if m→10 Sept 2026EmoMed: An Emotionally-Aware Agent for Multimodal Medical Support with Real-Time Information RetrievalarXiv:2609.07194v1 Announce Type: new Abstract: We present EmoMed - a multimodal medical consultation agent that adapts its responses based on users' emotional states while maintaining clinical accura→15 Sept 2026Efficiency Hallucination: Formalizing and Measuring Behavioral Calibration in LLM-Based Code OptimizationarXiv:2609.14839v1 Announce Type: cross Abstract: The integration of Large Language Models (LLMs) into automated code optimization introduces a critical reliability risk we term the Efficiency Halluci
CompanyMeta8 recent entries15 Sept 2026Mirror, Mirror on the Wall: Prompt Echoing in Small Instruct Language ModelsarXiv:2609.15045v1 Announce Type: cross Abstract: Prompt echoing is a recognized failure mode of instruct language models, in which a model instead of generating a response, mirrors the provided promp→16 Sept 2026Decoy Direction Optimization: A Post-Hoc Defense Against LLM AbliterationarXiv:2609.16204v1 Announce Type: cross Abstract: Safety guardrails in open-weight language models can be readily bypassed using Refusal Feature Ablation (RFA), a technique that identifies and project→18 Sept 2026The Neutral Mask: How Alignment Training Provides Shallow Alignment while Leaving Partisan Structure Intact in a Large Language ModelarXiv:2606.09735v2 Announce Type: replace Abstract: The ambition behind alignment training is to make large language models safe and useful. The primary mechanisms, reinforcement learning from human f→18 Sept 2026Safety Beyond the Interface: Detecting Harm via Latent States in Large Language ModelsarXiv:2609.19472v1 Announce Type: new Abstract: Autonomous systems increasingly rely on Large Language Models (LLMs) yet the safety infrastructure surrounding these models introduces latency and compu→21 Sept 2026HE-Guardrail: A Homomorphic Guardrail Against Jailbreak Attacks for Encrypted Large Language Model InferencearXiv:2609.21484v1 Announce Type: cross Abstract: Homomorphic encryption (HE) has emerged as a promising approach to privacy-preserving machine learning (PPML), enabling computation directly over encr→28 Sept 2026MoMHa: Multi-Objective Optimization of LLM Harnesses over Accuracy, Safety, and TokensarXiv:2609.30967v1 Announce Type: new Abstract: Most work on improving large language models treats accuracy as the sole objective. We argue that the harness, the Python code surrounding the model tha→1 Oct 2026Values as Style: Disentangling Values from Semantics with One-Way Mixing for Low-Damage LLM SteeringarXiv:2609.39701v1 Announce Type: new Abstract: Value steering should change an LLM's normative priorities while preserving the scenario, facts, and task constraints underlying its answer. Conventiona→2 Oct 2026Homomorphic Advantage Operator: Stabilizing Reinforcement Learning Under Fully Homomorphic Encryption ConstraintsarXiv:2610.02074v1 Announce Type: new Abstract: Privacy-preserving machine learning presents significant deployment challenges on the cloud for intelligent systems with confidential data. Fully Homomo
CompanyMistral8 recent entries27 Apr 2026TreeCoder: Systematic Exploration and Optimisation of Decoding and Constraints for LLM Code GenerationarXiv:2511.22277v2 Announce Type: replace Abstract: Large language models (LLMs) have shown remarkable ability to generate code, yet their outputs often violate syntactic or semantic constraints when →12 May 2026Less Diverse, Less Safe: The Indirect But Pervasive Risk of Test-Time Scaling in Large Language ModelsarXiv:2510.08592v3 Announce Type: replace-cross Abstract: Test-Time Scaling (TTS) improves LLM reasoning by exploring multiple candidate responses and then operating over this set to find the best out→18 May 2026Graph-Regularized Sparse Autoencoders for LLM Safety SteeringarXiv:2512.06655v3 Announce Type: replace-cross Abstract: Sparse autoencoders (SAEs) are increasingly used to extract activation directions for inference-time steering, but their standard sparsity obj→6 Jun 2026GuardNet: Ensemble Strategies of Shallow Neural Networks for Robust Prompt Injection and Jailbreak DetectionarXiv:2606.05566v1 Announce Type: new Abstract: Large Language Models (LLMs) have transformed natural language processing, but they remain vulnerable to Prompt Injection (PI) and Jailbreak (JB) attack→7 Jul 2026Governed MCP: Kernel-Level Tool Governance for AI Agents via Logit-Based Safety PrimitivesarXiv:2604.16870v2 Announce Type: replace-cross Abstract: AI agents increasingly call external tools (file system, network, APIs) through the Model Context Protocol (MCP). These tool calls are the age→27 Aug 2026A Layered Security Framework Against Prompt Injection in RAG-Based ChatbotsarXiv:2606.19660v2 Announce Type: replace-cross Abstract: Prompt injection is ranked as the most critical vulnerability in large language model (LLM) deployments by the OWASP Top 10 for LLM Applicatio→4 Sept 2026Measuring Harmfulness of Computer-Using AgentsarXiv:2508.00935v3 Announce Type: replace-cross Abstract: Computer-using agents (CUAs), which can autonomously control computers to perform multi-step actions, might pose significant safety risks if m→8 Oct 2026Quad-State Safety Evaluation of Open-Weight Large Language Models on Non-Canonical InputsarXiv:2610.09033v1 Announce Type: new Abstract: Standard safety evaluations of large language models assess harmful requests written in canonical plain text, while models in real-world deployment rout
CompanyxAI5 recent entries19 May 2026Multilingual jailbreaking of LLMs using low-resource languagesarXiv:2605.18239v1 Announce Type: cross Abstract: Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails. We investigate whether multi-turn conversation→23 Jun 2026BELLS-O: Evaluating the Operational Trade-offs of LLM Supervision SystemsarXiv:2606.20668v1 Announce Type: cross Abstract: LLM supervision systems, namely input/output moderation filters and jailbreak detectors, are the primary safeguard against misuse in deployed AI appli→10 Sept 2026XAI-Refine: An Automated Explanation-Knowledge Loop for Brain-Age PredictionarXiv:2609.09388v1 Announce Type: new Abstract: Brain-age prediction models are commonly evaluated by predictive accuracy, yet accurate predictions alone do not establish that a model relies on reprod→23 Sept 2026Decoding Guardrails: XAI-Guided Perturbation Analysis of Prompt Injection DetectionarXiv:2609.24801v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly deployed in production systems, raising concerns about their exposure to adversarial manipulation throug→25 Sept 2026LLM Agents Can Easily Tamper With Their Own TracesarXiv:2609.30266v1 Announce Type: cross Abstract: Asynchronous monitoring, incident investigations, and compliance audits primarily rely on agent traces to reconstruct what happened. These analyses as
CompanyDeepSeek8 recent entries30 Apr 2026Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning TrainingarXiv:2510.20956v2 Announce Type: replace-cross Abstract: We discover a novel and surprising phenomenon of unintentional misalignment in reasoning language models (RLMs), which we call self-jailbreaki→12 May 2026Scam2Prompt: A Scalable Framework for Auditing Malicious Scam Endpoints in Production LLMsarXiv:2509.02372v3 Announce Type: replace-cross Abstract: Large Language Models have become critical to modern software development, but their reliance on uncurated web-scale datasets for training int→19 May 2026Multilingual jailbreaking of LLMs using low-resource languagesarXiv:2605.18239v1 Announce Type: cross Abstract: Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails. We investigate whether multi-turn conversation→11 Aug 2026When Grammar Guides the Attack: Uncovering Control-Plane Vulnerabilities in LLMs with Structured OutputarXiv:2503.24191v4 Announce Type: replace-cross Abstract: Content Warning: This paper may contain unsafe or harmful content generated by LLMs that may be offensive to readers. Large Language Models (L→18 Aug 2026When Agentic Executions Fail: Detecting and Localizing Runtime Faults from TelemetryarXiv:2608.14680v1 Announce Type: new Abstract: Reliability in LLM-based agentic systems is a property of the whole execution (its tool calls, model calls, guardrails, and inter-agent messages), not o→3 Sept 2026FUSE: An Evaluating Framework for Dangerous Capabilities of LLMsarXiv:2609.02168v1 Announce Type: new Abstract: Fragmented safety evaluation undermines the governance of dangerous AI capabilities. We present a modular framework that evaluates each model through th→10 Sept 2026SRD-GUARD: A Defense Framework of LLMs via Semantic Rewriting and Joint Multi-Model Scoring for Latent Intent ExposurearXiv:2609.06540v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly deployed in safety-critical applications, yet jailbreak attacks can conceal harmful intent through role-→5 Oct 2026EvoRiskBench: An Evolving Benchmark for Runtime Security Risks in Workspace AgentsarXiv:2610.03153v1 Announce Type: cross Abstract: Workspace agents combine large language models with execution harnesses to perform stateful, multi-step tasks that access or modify external resources
CompanyNVIDIA2 recent entries21 Apr 2026CoDial: Interpretable Task-Oriented Dialogue Systems Through Dialogue Flow AlignmentarXiv:2506.02264v3 Announce Type: replace Abstract: Building Task-Oriented Dialogue (TOD) systems that generalize across different tasks remains a challenging problem. Data-driven approaches often str→30 Apr 2026Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning TrainingarXiv:2510.20956v2 Announce Type: replace-cross Abstract: We discover a novel and surprising phenomenon of unintentional misalignment in reasoning language models (RLMs), which we call self-jailbreaki