FreeLLMOpen-weight GLM‑5.3 builds full Firefox exploit chains without human input
Epoch AI shows the freely available model GLM‑5.3 can autonomously create multi‑stage exploits for Firefox versions in under 48 hours, costing about $1,000 per attack.
Researchers at Epoch AI demonstrated that the Chinese open‑weight model GLM‑5.3 can automatically generate full exploit chains against Firefox.
The model received only the target Firefox version, basic tools and public information, then identified upcoming security fixes, selected vulnerabilities and built multi‑stage attacks without any human guidance.
Across six Firefox releases from 17 September to 26 October 2026, the fastest successful exploit took 12 hours; all attacks were completed within a 48‑hour limit and each cost roughly $1,000 in API fees.
The exploits relied on a previously unknown vulnerability that the model rediscovered in every attempt, allowing it to bypass sandbox protections and read local files when a user visited a malicious webpage.
Mozilla patched the discovered vulnerability in Firefox 159.0 on 27 October 2026, but the researchers note the patch likely makes attacks “at least substantially more difficult.”
Ce que cela signifie pour vous
The result shows that freely accessible models can be weaponised to exploit the “patch gap” between fix publication and user update, raising concerns for all software vendors. Users of free or locally hosted AI should be aware that such models may be repurposed for offensive tasks, highlighting the need for robust safety controls and faster patch deployment.
Sources
- Open-weight model GLM 5.3 built full Firefox exploit chains with zero human input in under 48 hours, The Decoder, 30 Oct 2026
Rédigé automatiquement à partir des sources ci‑dessus, puis vérifié par notre code : chaque nombre et chaque nom doivent y figurer, et aucune phrase ne doit copier une source. Une erreur ? Signalez‑la.
Quelle IA pouvez‑vous utiliser gratuitement aujourd’hui ?
Tous les modèles IA gratuits, classés par votes humains anonymes, avec la mémoire et la vitesse dont ils ont besoin sur votre propre GPU.