Epoch AI finds that open-weight GLM 5.3 can exploit Firefox's patch gap for about $1,000 per version

Researchers at Epoch AI found that GLM 5.3, a Chinese open-weight model, can autonomously build Firefox exploits during the "patch gap," the period when a security fix is visible in the code but has not yet reached users, The Decoder reports. According to the report, the model exploited this gap across six Firefox versions released between September 17 and October 26. Each successful attack cost about $1,000 in API fees once the test setup was in place, and the report's headline says the chains were built in under 48 hours with no human input. The model identified security fixes that had been disguised in the code and chained those vulnerabilities with a previously unpublished flaw. The result bypassed browser protections and read local files when a user simply visited a webpage. Anthropic had previously reported that GLM 5.3 can write exploits, The Decoder notes. The researchers warn that similar patch-gap risks exist across the software industry. That matters because widely available models could shrink the time defenders have between publishing a fix and getting it installed, for any project whose patches are public before release.