GLM 5.3 Autonomously Exploits Firefox Patch Gaps
Epoch AI reports that the open-weight model GLM 5.3 autonomously chained vulnerabilities from published, not-yet-shipped Firefox security fixes into webpage-based exploit chains that read local files without further user interaction. The researchers successfully targeted all six Firefox versions released between September 17 and October 26, with the fastest attempt taking 12 hours and each successful attack using about $1,000 in compute after setup. Mozilla separately patched the undisclosed vulnerability in Firefox 159.0, but the researchers did not test that release, and similar attempts against Google Chrome failed after about $100,000 in compute spending.
The reported experiment shows that a widely available open-weight model can turn public patch information into working browser exploits at comparatively low cost, making patch deployment timing a direct AI-security concern.
This source provides an excerpt. Continue reading on the original site.
#GLM 5.3#Z.ai#Firefox#Mozilla#Google Chrome#Kimi K3.1 Pro#Anthropic#Mythos