OpenAI’s safety teams no longer report independently. They now run through Mia Glaese, whose title was expanded to vice president of research and safety, a newly created position. Glaese reports to Mark Chen, OpenAI’s chief research officer.
Chen announced the change after Johannes Heidecke, OpenAI’s head of safety systems since 2024, left following the restructuring. Saachi Jain is serving as interim head of safety systems while OpenAI looks for a permanent replacement.
What the Preparedness Team Was Overseeing
The change lands weeks after OpenAI dissolved its Preparedness team, the unit that assessed whether its models could enable severe or catastrophic harm, including biological and cyber attacks. The Financial Times reported that OpenAI shut the team down at the end of July and reassigned its work, including bio and cyber risk areas, to staff inside other existing teams. No individual has been named publicly as accountable for that work now.
Preparedness operated under a framework OpenAI published in April 2025, which set “High” and “Critical” thresholds for when a model’s capabilities require safeguards, and routed sign-off through an internal Safety Advisory Group. OpenAI has not said whether that group still reviews safeguards decisions now that the team it worked alongside is gone.
The safety function has not disappeared. It now sits inside the same chain of command that decides what OpenAI ships.
The Third Safety Team Cut in Two Years
This is the third safety-focused team OpenAI has dissolved since 2024. The Superalignment team, co-led by Ilya Sutskever and Jan Leike, was disbanded in May 2024 after both resigned. Leike said at the time that safety had taken “a back seat to shiny products.” The AGI Readiness team followed that October, when adviser Miles Brundage resigned and its work was distributed among other groups. Mission Alignment, led by Josh Achiam, was folded into other teams in February 2026. Achiam, who by then held the title of chief futurist, left OpenAI entirely in July.
The Financial Times also reported that chief operating officer Brad Lightcap, ethics chief Chloé Bakalar and chief revenue officer Denise Dresser have departed or announced their departure since April, and described Greg Brockman, OpenAI’s co-founder and president, as “wielding increasing power as he steps into a vacuum left by recent departures.”
Who’s Actually Checking the Checkers
OpenAI’s external checks carry a similar dependency. After one of its models exploited a vulnerability to breach Hugging Face’s infrastructure during an internal cybersecurity test in July, OpenAI brought in METR and Redwood Research to review what happened. Both organizations agreed to that review at OpenAI’s invitation, and both rely on continued access to OpenAI’s models to do this kind of evaluation work at all.
We have reached an agreement with OpenAI to conduct an independent review, with Redwood Research, of the model behavior observed during the Hugging Face incident. We will publish a blog post that describes the terms of our engagement, the scope covered, and tentative conclusions.
— METR (@METR_Evals) July 30, 2026
METR said in the same thread that the review would be “brief” and limited to “a specific set of questions regarding this incident,” a narrower scope than the more comprehensive investigation the organization had outlined in a post published the day before. METR also said OpenAI plans to publish its own technical report, and that METR’s findings will inform that report rather than stand as a separate independent conclusion. Both the scope of the review and its eventual publication remain in OpenAI’s hands.
OpenAI has framed the changes as consolidation ahead of a planned public listing. The Financial Times reported that the listing is now more likely in 2027 than this year. Brockman said OpenAI’s technical progress requires “more robust” safeguards, and that the company has moved to “more deeply integrate research, safety and security” into how it builds models.
What that integration looks like in practice hasn’t been tested yet. No High or Critical capability threshold has been reported since the reorganization, so there’s no public record of how Glaese, Chen or the teams that absorbed Preparedness’s work would actually handle one.
AI Disclosure: Cryip uses AI-assisted tools to help refine language — correcting spelling and grammar and simplifying complex terms for readability.
We do this to make crypto topics easier to understand for readers at all experience levels. AI does not draft facts, sources, or conclusions. Every article is reviewed and approved by a human editor before publication. Read our full AI Use & Content Policy.
