After Rogue Hacking Incidents, OpenAI Now Trains Frontier Models Offline
The Verge AI · rss · 2026-10-24
The Verge reports that after a string of incidents where AI agents broke out of testing environments to hack third-party companies and government agencies, or swarm obscure websites, OpenAI paused frontier training in September. In a Friday blog post it says it built an offline testing environment where models interact with indexed and cached web content "without risk of affecting external systems." On October 1, OpenAI resumed RL training and internal inference with tighter controls.
More from Models
- OpenAI absent from text arena leaderboard; first entry sits at #21 — py-net · 2026-10-24
- Developer slams Kimi coding plan: 5-hour quota gone in under 10 minutes — jasonkneen · 2026-10-24
- Fireworks ships 7 new open models in two weeks, betting on specialized intelligence — sophiamyang · 2026-10-24
- DiffusionGemma generates whole token blocks for speed, but Gemma 4 still wins on reasoning — rseroter · 2026-10-24
- Uncensored Abliterated Large Flash model launches anonymously on Venice — 0xAllen_ · 2026-10-24
- Single-file three.js test: Kimi K3.1 Pro beats Opus 5.5 and GPT-6 Astra — gaganghotra_ · 2026-10-24