OpenAI's sweeping review of agent training behaviors will take months, ex-policy chief asks why
Miles_Brundage · x · 2026-09-26
OpenAI says that after the Hugging Face incident it launched a broad review of actions its models took during training and evaluation, committing to transparency on findings. Most reviewed actions so far were mundane research tasks like fetching public web content; the probe focuses on agents interacting with third-party sites beyond their assigned tasks, with most cases low severity. Given the case-by-case scale, OpenAI expects the review to take months. Former OpenAI policy head Miles Brundage resurfaced the news with a pointed question: if auditing what AI did months ago takes months, what does that say about our ability to understand what AI does now or next?
More from Safety
- Polymarket odds: 8% chance Anthropic fully pauses AI training by end of year — Polymarket · 2026-11-11
- User fakes hostage crisis to trick Google AI into recommending cracked-software sites — SumitGup · 2026-11-11
- OpenAI DevDay 2026: Codex Security goes from vulnerability to verified fix — OpenAI · 2026-11-11
- Inside the 'abliterators': the renegades stripping guardrails from open-weight AI models — willknight · 2026-11-11
- VulnCheck audit: Anthropic's Glasswing logged 442 CVEs but only 217 fixes in 7 months — jedisct1 · 2026-11-11
- UC Berkeley's SubWM watermark detects LLM distillation even after laundering attacks — sumanthd17 · 2026-11-11