Open-weight AI model GLM 5.3 turns Firefox security fixes into working exploits for about $1,000
Open-weight AI models can exploit the patch gap
Researchers at Epoch AI found that GLM 5.3, a freely available open-weight model from Z.ai, could read upcoming Firefox security patches and build working exploits against all six releases between September 17 and October 26, 2026. Visiting a single webpage let an attacker exfiltrate local files with no user interaction. Each compromise cost roughly $1,000 in API calls, and no provider safeguards blocked the work. Mozilla has patched an additional undisclosed flaw in Firefox 159.0.
Our central finding is that widely available open-weight models can now do much, or even all, of this work on their own.