Google's CodeMender agent upstreams 72 reviewed security fixes
Google DeepMind reported that CodeMender had upstreamed 72 security fixes to open-source projects, including codebases as large as 4.5 million lines. The agent used recent Gemini Deep Think systems to discover and repair vulnerabilities, with human review required before patches were submitted.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
The upstreamed patches establish verified practical impact beyond a benchmark or forecast. The result increases defensive software capacity, while human review limits autonomous external effects. No exact public model version was identified, so the item does not claim a model relationship.
Assessment history
-
R1
Away 36 · confidence 88
New practical-impact safety evidence found in the October 2025 gap review.
13 Aug 2026
Share this page
-
DoomBench assesses “Google's CodeMender agent upstreams 72 reviewed security fixes” as evidence moving away from doom, with magnitude 36 and confidence 88 out of 100 in the safety and alignment category.
-
The DoomBench assessment of “Google's CodeMender agent upstreams 72 reviewed security fixes” is based on reporting from Google DeepMind and records the editorial rationale, source quality, attribution, and revision history.
-
DoomBench summarizes “Google's CodeMender agent upstreams 72 reviewed security fixes” as follows: Google DeepMind reported that CodeMender had upstreamed 72 security fixes to open-source projects, including codebases as large as 4.5...
https://www.doombench.com/news/google-s-codemender-agent-upstreams-72-reviewed-security-fixes-2025-10-06