Open-Weight GLM 5.3 AI Autonomously Creates Security Exploits for 6 Recent Firefox Versions

The open-weight AI model GLM 5.3 autonomously created security exploits for six recent Firefox versions by targeting the gap between public security fix announcements and their deployment, according to tests by Epoch AI. Each exploit took 12 to 48 hours to develop at a cost of approximately $1,000 and allowed researchers to steal browser cookies, passwords, and local files without user interaction. The model also discovered a previously unknown Firefox vulnerability that Mozilla patched in its Firefox 159.0 release on October 27.