Epoch researchers used GLM 5.3 to exploit six Firefox releases

The file-reading attacks combined clues in published security fixes with an unpublished flaw; Mozilla patched that flaw in Firefox 159, though Epoch has not tested whether the update blocks the attack.

By · Published

Primary source: Hacker News

Why it matters

The Firefox tests show that public security fixes can give an open-weight model enough information to build a working file-reading exploit before users receive the patch. Epoch's $1,000 estimate applies only after its harness was built, and Firefox 159's effectiveness against the full attack path remains unconfirmed.

A gloved hand pulls a paper from a file organizer beside a laptop showing the Firefox logo, with “z.ai” signage and an “RW” mug behind it.

Epoch AI researchers used Z.ai's open-weight GLM 5.3 model to build file-reading exploits against six Firefox releases, turning public security fixes into clues about vulnerabilities before those fixes reached users. The experiments show how quickly the "patch gap" can become a practical target: after building its test setup, Epoch estimated each successful attack cost about $1,000 in model API fees.…

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Reader comments

Conversation for this story loads after sign-in.