GLM 5.3 builds Firefox exploit chains autonomously

Epoch AI researchers demonstrated that the open-weight AI model GLM 5.3 can autonomously build full Firefox exploit chains with zero human input. The model successfully exploited Firefox before publicly visible security fixes reached users. It identified disguised security fixes and chained vulnerabilities with a previously unpublished flaw to bypass browser protections. Once the test setup was in place, exploits cost about $1,000 in API fees per Firefox version. The fastest successful attack took 12 hours, while researchers gave the model a maximum of 48 hours.

The model built a webpage that could read local files from a visitor's computer and send them to an attacker's server when the page was loaded. GLM 5.3 relied on a previously unknown vulnerability that it independently rediscovered in every attempt to bypass Firefox's sandbox protections. Mozilla patched this vulnerability in Firefox version 159.0, released on October 27, 2026. Using the same setup against Google Chrome, the team spent about $100,000 in API fees without producing an equivalent exploit.

Neither Z.ai nor other API providers triggered any safety guardrails or refusals when researchers used GLM 5.3 for exploit development. The researchers recommend keeping security fixes private until they are ready to ship and moving them from beta to general release more quickly. Firefox and Google Chrome have already begun shipping weekly updates to help reduce patch gap risks. Researchers caution that better elicitation methods or more capable models might still succeed in exploiting Chrome.