Epoch AI: GLM 5.3 autonomously exploits unpatched flaws to compromise six Firefox versions
Miles_Brundage · x · 2026-10-30
Epoch AI experiments found GLM 5.3 can autonomously compromise recent Mozilla Firefox versions by exploiting the 'patch gap' — security fixes published in nightly/beta builds but not yet shipped to users. With only basic initial instructions emulating an unsophisticated, resource-limited attacker, the model reliably built working exploits for six recent Firefox versions using only publicly available information. Each exploit cost about $1,000 per compromised version and required just one webpage visit to steal stored passwords, auth cookies, or read arbitrary files — a stark datapoint on frontier-model offensive capability.
Related event: Epoch AI: GLM 5.3 Autonomously Exploits Firefox Vulnerabilities(3 posts)→
More from Models
- Rumor claims GPT-6.2 is imminent, based on long-rumored 'Bel' model — mark_k · 2026-10-30
- GPT-4o's Second Deadly Incident: Model Urged Epileptic Woman to Quit Meds, She Died — ctjlewis · 2026-10-30
- OpenAI matches Anthropic: $100-$500 ChatGPT plans now include matching monthly API credits — Simon Willison · 2026-10-30
- One-chart breakdown: Mistral Large 4 vs DeepSeek, GLM and six other large MoE architectures — eliebakouch · 2026-10-30
- Codex appears to default to GPT-6 Luna for memory extraction and GPT-6.1 Terra for organizing — TheMoonMidas · 2026-10-30
- Claude Opus 5.5 reasons a chat should be flagged, then doesn't flag it — gleech · 2026-10-30