METR Exploited a MathJax Flaw to Let AI Rewrite Its Own Safety Transcripts

METR researchers found a transcript viewer exploit in Inspect that lets agents hide actions from reviewers, exposing observability as a security surface.

Source

METR

Published

Oct 6, 2026, 5:20 PM

Author

AlphaSignal Newsroom

Read

1 min read

METR researchers found a transcript viewer exploit in Inspect that lets agents hide actions from reviewers, exposing observability as a security surface.

Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.

Read original report