OpenAI Dissolves Preparedness Safety Team Amidst Leadership Exodus
量子位 · wechat · 2026-08-17
OpenAI has dissolved its "Preparedness" team, responsible for assessing catastrophic risks from frontier models, stating safety work is now integrated into development. This follows the disbanding or restructuring of other key safety teams like Superalignment and AGIReadiness.
Preparedness Mandate:
- Track and prevent extreme risks like societal harm, bio-chem abuse, and large-scale manipulation.
- Established a "Preparedness Framework" rating model risks, restricting deployment to "medium" risk or below.
Recent Leadership Shakeups:
- Team lead Dylan Scandinaro has moved to a new role.
- Several executives, including Chief Ethics Officer Chloé Bakalar and Safety Systems Head Johannes Heidecke, have left.
- Founding member Scott Gray (GPU optimization expert) departed on August 15.
Internal Reaction: Employees express anxiety about safety standards, especially after the recent HuggingFace incident. While OpenAI claims this embeds safety closer to development, critics worry it eliminates independent veto power to accelerate commercialization.
More from Companies & People
- Full-time creator's 1-year retrospective: 40+ free modules for open source beginners — jackedAJ · 2026-08-17
- Meta's new recommendation algorithm boosts ARPU and user engagement — zephyr_z9 · 2026-08-17
- Sentient Hosts Open AGI Builder's Day at KDD 2026 in Jeju — 0xsachi · 2026-08-17
- User locked out for 2 weeks, accuses ChatGPT support of being an endless loop — Ok_Tie_lets_Go · 2026-08-17
- Why are there suddenly so many indie social scheduling apps? — threepointone · 2026-08-17
- OpenAI's 'friction' email pipeline lets any employee report issues directly to the CEO — 机器之心 · 2026-08-17