What's this about?
A report that OpenAI dissolved the team responsible for assessing catastrophic model risks, framed as streamlining ahead of an IPO.
What exactly did the Preparedness team do?
Preparedness was the function that measured how dangerous a model was before release — running evaluations in domains like cyber and bio, and determining which capability threshold a model had reached under the Preparedness Framework. It's the measurement arm behind the commitments: the framework says "we won't deploy past this line without these safeguards," and Preparedness is what tells you whether you're past the line. Disbanding a team is not the same as abandoning the function, which may have been redistributed — but the framework's credibility depends on somebody independent doing that measuring.
Why is this worth noticing?
HN 31 points, 14 comments
- The juxtaposition is the story. This surfaced the same week OpenAI paused its largest frontier RL run explicitly because a Preparedness Framework threshold might have been crossed — the framework doing exactly what it's for.
- "IPO streamlining" as the stated reason connects to the broader pattern several commenters have flagged this month: safety posture and public-market pressure pulling in different directions.
- Read carefully. A team being dissolved may mean the work moved rather than stopped, and the reporting is thin. It's worth watching rather than concluding from.