Epoch AI Experiment: GLM 5.3 Independently Compromises Multiple Firefox Versions by Exploiting Patch Gaps

An Epoch AI experiment found that the open-weight model GLM 5.3, released by Z.ai in August, can reverse-engineer vulnerabilities from Mozilla security fixes that were published but not yet pushed. Using approximately $1,000 of compute power, it independently compromised all six Firefox versions released between September 17 and October 26. The fastest attack took only 12 hours, achieved by chaining patch-gap vulnerabilities with undisclosed flaws to bypass protections.