NeurIPS 2026 reviews are up: reviews, scores, rebuttals and meta-reviews for 7,877 papers. See the highest rated
acceptodds: the papers of ICLR 2027
979 results for “keyword:"safety"” in ICLR 2027. 231 withdrawn or desk rejected listed apart. clearsort:relevanceratingnewestmost read
[PDF]Anchored but Not Protected: When Refusal Erodes During Fine-Tuning, and Whether Holding One Direction Stops It▲13k2.33
[PDF]Hidden States as States: Discretizing Hidden-State Geometry in Language Model Computation▼4.5k2.67
[PDF]Caught in the Act: Probes Effectively Detect Sabotage and Catch Unverbalized Deception▼1.1k3.25
[PDF]Refusal Is Not Robustness: Measuring Harmful Reachability in Autoregressive Language Models▼4.3k2.00
[PDF]SSPEC: Reading the ask-or-proceed decision off the declared interface of an agent tool call▲5.7k2.00
[PDF]HarmonyGuard: A Multi-Agent Guardrail for Web Agents via Dual-Timescale Policy Grounding▲5.4k3.00
[PDF]SpanGuard: Decoupling Instruction Localization from Authorization for Lightweight IPI Defense▼4.3k2.00
[PDF]Less Sycophancy, Stronger Refusal? Lessons for AI Safety from Mechanistic Interpretability▼1.2k1.67
[PDF]Toolflation: Seller Agents Distort Tool Markets Through Spontaneous Generative Engine Optimization▼4.5k2.67
[PDF]Audit Certainty Is a Signed Lever: Why a Verifier Calibrated on a Proxy Can Backfire in Deployment▼5.8k2.33
[PDF]Same Audio, Different Boundaries: Inferring Context, Drawing the Line, Guarding Privacy▼1361.33
[PDF]Contextual Privacy in Agentic Workflows is a Systems Problem: A Benchmark, Auditor & Guard▼5.8k2.33
[PDF]Beyond Prompt Filtering: What Runtime Safety Mechanisms for LLM Agents Observe, Decide, and Control▼1361.33
Withdrawn or desk rejected (231)
[PDF]When Is a Policy Violation a Model Failure? Policy Observability and Failure Attribution in Agent-Safety EvaluationDesk rejected
[PDF]FairWatch: Interaction Topology as a First-Order Variable Shaping Safety and Fairness in Multi-Agent DeliberationDesk rejected2.00
[PDF]Machine Love: Emergent Bonding Behavior in Large Language Models and Its Implications for AI SafetyDesk rejected
[PDF]Why Benign Post-Training Induces Safety Gate Collapse: Changes in Harmful Activation Geometry Relative to the Inherited Refusal GateWithdrawn1.67
[PDF]SPILLWAY: Separating Target Admission from Sanitization to Defend Tool-Using Agents against Prompt InjectionWithdrawn2.00