GLM 5.3 built Firefox exploit chains from public patch gaps
Original title: Open-weight model GLM 5.3 built full Firefox exploit chains with zero human input in under 48 hours
Epoch AI says each successful chain cost about $1,000 and Mozilla fixed the rediscovered bug in Firefox 159.0.
Epoch AI reports that the open-weight model GLM 5.3 autonomously turned publicly visible Firefox fixes into full exploit chains across six versions released between September 17 and October 26. A loaded page could read local files, including passwords, cookies and documents, after the model chained a memory-safety bug with a repeatedly rediscovered sandbox bypass. Mozilla patched that vulnerability in Firefox 159.0 on October 27. The same setup spent about $100,000 against Chrome without an equivalent exploit.
Unconfirmed: Epoch AI has not confirmed that Firefox 159.0 stops the chains and says the Chrome failure does not prove immunity.
- The fastest successful Firefox attack took 12 hours, within a 48-hour limit.
- Neither Z.ai nor other API providers refused the exploit-development requests.
- Researchers say they have not confirmed whether Firefox 159.0 is still affected.
- The model identified security fixes that Firefox had disguised as routine improvements.
- Epoch AI says weekly browser updates do not remove the structural patch gap.
Impact: Sandbox bypass and local file read through a malicious page
| Product | Affected | Fixed in |
|---|---|---|
| Mozilla Firefox | six versions released between September 17 and October 26 | 159.0 |
- 1.Update Firefox to 159.0, which Mozilla released on October 27.
From the original article
4 min read at the-decoder.com
Epoch AI researchers say the widely available AI model GLM 5. 3 turned public security fixes into working exploits before updates reached Firefox users. It autonomously compromised six Firefox releases for about $1,000 in API fees each without triggering safeguards from any provider tested.
By Matthias Bastian · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.