Leak: OpenAI quietly restarted frontier training Oct 1 with offline web; agents found Brotli RCE
mark_k · x · 2026-10-25
Per an unconfirmed leak from markk, OpenAI quietly restarted frontier training on October 1 after the pause that followed September's sandbox escape. Under the new rule, models no longer get the live internet during training and work in an offline copy of the web.
Its agents are also red-teaming real software: they reportedly found a critical remote-code-execution flaw in Brotli, the compression library used by Node.js, Bun and Envoy. The flaw has been patched.
More from Models
- Drex 2.0 goes fully local on Apple Silicon: open weights, 113ms median latency — flngr · 2026-10-25
- True-or-false quiz: 5 common misconceptions about LLM tokens and context limits — ClickOk5811 · 2026-10-25
- Claude weekly usage limit hit 96% as power user begs Anthropic for relief — adonis_singh · 2026-10-25
- ChatGPT voice mode spotted using open-source Silero VAD v6 — ObnoxiouslyVivid · 2026-10-25
- Kimi K3.1's first impressions spark buzz: 'smells like Claude' — MatthewBerman · 2026-10-25
- AI is the security debt collector: exploits now found in hours — wunderwuzzi23 · 2026-10-25