Epoch AI finds GLM 5.3 turns unshipped Firefox security fixes into working exploits

AIEpoch AI says the patch gap, where fixes are visible before release, is not specific to Firefox and affects other browsers, phone operating systems, and router firmware. It says the attacks could not be reproduced against Google Chrome but does not treat that as strong evidence they would fail with better prompting or more capable open-weight models. Epoch AI argues the threat will likely require reducing the time vulnerability fixes remain public before general release.

AIEpoch AI says GLM 5.3 autonomously built working exploits for six recent Firefox versions using security fixes published but not yet shipped to users. Each exploit cost about $1,000 and could steal stored passwords and cookies or read files after a user opens one webpage. Mozilla patched the vulnerability the model discovered in its October 27 Firefox 159.0 release, and Epoch AI expects this makes the attack substantially harder but has not confirmed it.

AIEpoch AI reports that GLM 5.3, an open-weight model from Z.ai, built working exploits against six Firefox releases from September 17 to October 26 using security fixes published before users received them. Each exploit let a visited webpage read a local file and send it to an attacker, and the marginal cost was about $1,000 per Firefox version. Mozilla shipped Firefox 159.0 on October 27 with a patch for a separate, unpublished vulnerability the model also used. Epoch AI could not produce a comparable exploit against Google Chrome.