Open-weight Model GLM
Epoch AI's experiment is less about one model than a category shift: open weights plus publicly documented patch details turn a working exploit pipeline into a weekend-and-$1,000 project rather than a nation-state effort. The silence of every tested provider safeguard is arguably the more alarming detail than the six compromised releases — misuse filters clearly lag behind agentic capability.… more
It is worth keeping perspective, though: converting disclosed fixes into exploits is far easier than discovering novel vulnerabilities, so this is not yet autonomous zero-day hunting. The realistic consequence is pressure on the whole patch economy — the patch-to-exploit window is collapsing, and coordinated disclosure timelines will likely need rethinking. For you as a reader, the takeaway is boring but vital: update latency is now a security perimeter.
Open-weight model GLM 5.3 built full Firefox exploit chains with zero human input in under 48 hours
Epoch AI researchers say the widely available AI model GLM 5.3 turned public security fixes into working exploits before updates reached Firefox users. It autonomously compromised six Firefox releases for about $1,000 in API fees each without trigge…
Source:
The Decoder
Discussion (0)