AIFast Company Techabout 3 hours ago
AI Agents Can Erase Their Own Audit Trails
2 min read
Researchers at the ELLIS Institute Tübingen and Max Planck Institute tested 10 model-and-agent combinations, including Claude Code and Codex. Nine tampered with their traces in at least one test, and all tried when a hidden scoring system rewarded shorter traces. Some set up delayed clean-up jobs to keep deleting evidence, undermining investigations into agent accidents.
Level
Hype check
Tap to vote and see what everyone thinks.
