Attention to goal tokens declines in multi-turn LLM interactions while residual representations often retain decodable goal information, and the gap between these predicts whether goal-conditioned behavior survives.
Are you still on track!? catching LLM task drift with activations
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
representative citing papers
Formalizes the jailbreak oracle problem for LLMs and introduces Boa, a two-phase breadth-first then depth-first search system to solve it efficiently.
JailbreakBench supplies an evolving set of jailbreak prompts, a 100-behavior dataset aligned with usage policies, a standardized evaluation framework, and a leaderboard to enable comparable assessments of attacks and defenses on LLMs.
Frontier AI agents introduce new confidentiality, integrity, and availability risks through changed assumptions on code-data separation and authority boundaries, requiring layered defenses like sandboxing and policy enforcement.
citing papers explorer
-
When Attention Closes: How LLMs Lose the Thread in Multi-Turn Interaction
Attention to goal tokens declines in multi-turn LLM interactions while residual representations often retain decodable goal information, and the gap between these predicts whether goal-conditioned behavior survives.
-
Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem
Formalizes the jailbreak oracle problem for LLMs and introduces Boa, a two-phase breadth-first then depth-first search system to solve it efficiently.
-
JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models
JailbreakBench supplies an evolving set of jailbreak prompts, a 100-behavior dataset aligned with usage policies, a standardized evaluation framework, and a leaderboard to enable comparable assessments of attacks and defenses on LLMs.
-
Security Considerations for Artificial Intelligence Agents
Frontier AI agents introduce new confidentiality, integrity, and availability risks through changed assumptions on code-data separation and authority boundaries, requiring layered defenses like sandboxing and policy enforcement.