Open-weight GLM 5.3 autonomously built full Firefox exploit chains in under 48 hours for ~$1,000
The Decoder · rss · 2026-10-30
Epoch AI researchers report that widely available open-weight model GLM 5.3 autonomously turned public security fixes into working Firefox exploits before updates reached users—no human input, under 48 hours, compromising six Firefox releases at $1,000 in API fees each, without tripping safeguards from any provider tested. The result raises serious concerns about open-weight models enabling automated offensive security.
More from Models
- Xiaomi open-sources Xiaomi-OCR-0, a 0.8B document understanding model topping sub-1B parsing — gnukeith · 2026-10-30
- Unsloth says it can't reproduce GSQ-RCO results, ships DeepSeek-V4.1 Dynamic v3.1 GGUFs — danielhanchen · 2026-10-30
- OpenAI's subscription pain: low limits and unbanked resets drag utilization — teortaxesTex · 2026-10-30
- Subscription utilization test: Claude Max hits 80-90%, OpenAI only ~50% — teortaxesTex · 2026-10-30
- Unsloth rebuts GSQ-RCO: DeepSeek-V4.1-Flash 1bit AIME25 is 91.88%, not 100% — danielhanchen · 2026-10-30
- SAM3 now runs offline on your GPU: ONNX export hits 0.37s on M5, 0.08s on RTX 3080 Ti — CSProfKGD · 2026-10-30