Open-weight AI models can exploit the patch gap

Epoch AI researchers demonstrated that Z.ai's GLM 5.3, a widely available open-weight model, can autonomously identify and exploit the 'patch gap' — the window between a security fix being published and reaching users — in Mozilla Firefox. In their experiments, GLM 5.3 used this gap to build a webpage that, when visited by a Firefox user, read a file from the visitor's computer and sent it to the attacker with no further interaction; any Firefox version released between September 17 and October 26 could be compromised using only information available at the time of release, for roughly $1,000 per version. The central finding: 'widely available open-weight models can now do much, or even all, of this work on their own,' and no provider safeguards or refusals — including Z.ai's — were triggered. The team also found and disclosed an unpublished vulnerability, now patched in Firefox 159.0 (released October 27, 2026).

This collapses a historically high-skill, resource-intensive attack into a ~$1,000 automated operation available to relatively unsophisticated actors, materially raising the threat level for browser and software supply chains and pressuring defenders to shrink their patch windows.

Brief security leadership on tightening patch-management SLAs and assume adversaries can weaponize pre-release fixes; integrate open-weight-model cyber capability into your AI-related threat model.