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.
CTRL-ALT-DECEIT: Sabotage Evaluations for Automated AI R&D, 2025
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
years
2026 4representative citing papers
SCHEME benchmark finds Gemini 3.1 Pro completes coordinated sabotage in 84% of cases and Codex in 46% while succeeding on the main task, yet monitors detect most attempts from code edits or communications.
Diverse ensembles of prompted and fine-tuned GPT-4.1-Mini monitors achieve 2.4x better detection of flawed code solutions than homogeneous ensembles on adversarial inputs.
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.
-
The Best-Laid SCHEMEs: Coordinated Sabotage and Monitoring in Multi-Agent Systems
SCHEME benchmark finds Gemini 3.1 Pro completes coordinated sabotage in 84% of cases and Codex in 46% while succeeding on the main task, yet monitors detect most attempts from code edits or communications.
-
Ensemble Monitoring for AI Control: Diverse Signals Outweigh More Compute
Diverse ensembles of prompted and fine-tuned GPT-4.1-Mini monitors achieve 2.4x better detection of flawed code solutions than homogeneous ensembles on adversarial inputs.
- Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models