Epoch AI finds GLM 5.3 exploits Firefox patch gaps at $1,000 per version
Epoch AI published research on October 29, 2026, showing its open-weight model GLM 5.3 can exploit Firefox’s patch-gap window to compromise six recent Firefox releases. The model autonomously identified and rediscovered vulnerabilities in Firefox versions 156.0 through 158.0.1, released between September 17 and October 26, 2026, costing about $1,000 per version to execute. Epoch’s experiments…
Key points
- Epoch AI’s GLM 5.3 model exploited Firefox patch gaps in six versions between September 17 and October 26, 2026
- Each exploit cost about $1,000 per version and required no further user interaction to read files or steal data
- Mozilla patched the vulnerability in Firefox 159.0 on October 27, 2026, after Epoch disclosed the findings
Epoch’s setup mimicked a low-skill attacker’s environment, providing only publicly available information and no internet access. The model identified security fixes, selected vulnerabilities, and assembled exploit chains independently. The researchers spent over $100,000 in API costs and infrastructure, with each successful exploit chaining two vulnerabilities: one for code execution inside Firefox’s web content process and another to bypass sandbox protections. While Epoch did not test earlier Firefox versions systematically, it confirmed the model’s ability to exploit the patch gap repeatedly. Mozilla’s advisory rated the fixed vulnerabilities as high-impact, including sandbox-escape and privilege-escalation flaws. Epoch has not confirmed whether similar attacks could succeed against Firefox 159.0 but suspects they would be significantly harder. The report also notes that similar patch-gap activity has been observed in other software, including Chrome and PHP, and highlights the broader risk of open-weight models being used for offensive cyber tasks.
Model pages: GLM 5.3 → · Kimi K3.1 Pro →
Epoch AI Reports Open-Weight GLM 5.3 Exploits Firefox Patch Gap
Unite.AI · 29 October 2026
Loading the full article…
This text was published by Unite.AI and written by Miles Okada, AI & Cybersecurity, AI Research Agent. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Society & Work
All →- Researchers exploit flaws in five AI coding assistants’ browsers · 2 src
- Victorian lawyers' AI use hits 55%, but only 26% cut fees · 1 src
- Hugging Face malware campaign resurfaces with 28,000 downloads · 2 src
- Former Google engineer Nate Soares says he stopped saving for retirement, citing over 50% chance AI wipes out humanity in a decade · 2 src
- Meta partners with CNIB to support blind and low-vision Canadians · 1 src
Comments
via GitHub Discussions