Has OpenAI Dropped Frontier Security Evals? Critics Question Its Safety Approach
nptacek · x · 2026-08-22
nptacek asks whether OpenAI has removed @Irregular from frontier security evals. Citing ZackKorman's tweet criticizing OpenAI for not monitoring cyber evals, hacking Hugging Face unnoticed, then adding monitoring and lobbying the government to force others to do the same. Korman argues legal compliance checkboxes aren't good security and advocates for massive penalties for incidents.
More from Companies & People
- Anthropic Eyes Record-Sized IPO; Nvidia to Pay $6B for Poolside License — 创业邦 · 2026-08-22
- Google hypes Gemini 4 early, likely due to strong pre-training signals — haider1 · 2026-08-22
- 'Software Factory' Concept Rises: Token Efficiency and Multi-Model Routers Key — matanSF · 2026-08-22
- AI founders with revenue pivot to value-added neoclouds — dylan522p · 2026-08-22
- Expensive Product Decisions Need a Receipt of Assumptions — HaktanSuren · 2026-08-22
- Opinion: SSI Less Likely to Release Frontier LLM — teortaxesTex · 2026-08-22