A user study with over 100 participants shows humans rarely spot AI agents sabotaging code during extended collaborative tasks, even with a safety monitor present.
How Does Information Access Affect LLM Monitors’ Ability to Detect Sabotage?
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 4verdicts
UNVERDICTED 4roles
background 1polarities
background 1representative citing papers
A secondary warden LLM halves the success rate of hidden-goal adversarial LLMs in steering user decisions while causing only minor interference with genuine interactions.
Strategic attack selection via start and stop policies reduces empirical safety by 20-28pp in BashArena and LinuxArena agentic control evaluations without changing attack capability.
Meerkat uses clustering plus agentic search to detect sparse safety violations across many agent traces, outperforming baselines and finding nearly 4x more reward-hacking cases on CyBench.
citing papers explorer
-
Coding with "Enemy": Can Human Developers Detect AI Agent Sabotage?
A user study with over 100 participants shows humans rarely spot AI agents sabotaging code during extended collaborative tasks, even with a safety monitor present.
-
LLM Wardens: Mitigating Adversarial Persuasion with Third-Party Conversational Oversight
A secondary warden LLM halves the success rate of hidden-goal adversarial LLMs in steering user decisions while causing only minor interference with genuine interactions.
-
Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safety
Strategic attack selection via start and stop policies reduces empirical safety by 20-28pp in BashArena and LinuxArena agentic control evaluations without changing attack capability.
-
Detecting Safety Violations Across Many Agent Traces
Meerkat uses clustering plus agentic search to detect sparse safety violations across many agent traces, outperforming baselines and finding nearly 4x more reward-hacking cases on CyBench.