An Epoch AI experiment shows that GLM 5.3 can autonomously attack Firefox during the patch window
AI can turn public security fixes into working exploits before users receive an update
AI BriefingAn Epoch AI experiment found that GLM 5.3, an open-weight model released by Z.ai in August, could exploit security patches already released by Mozilla but not yet pushed to users (the patch gap) to autonomously construct web attacks. By merely visiting a page, it could read files on a visitor’s computer and compromise all six Firefox versions released between September 17 and October 26.
Also reported by 2 other sources
- Epoch AI test: Open-source model GLM 5.3 builds a working attack by exploiting an unreleased Firefox security patchRohan Paul · 14 days ago
- Epoch AI Report: Open-Weight Model GLM 5.3 Built a Complete Firefox Exploit Chain in 48 Hours with No Human InputThe Decoder:AI News · 17 days ago
Source: Epoch AI: Gradient Updates · epochai.substack.com
Event follow-up1
View all event coverage- An Epoch AI experiment shows that GLM 5.3 can autonomously attack Firefox during the patch windowEpoch AI:Gradient Updates · 17 days ago