Analysis
OpenAI fired three researchers from its safety organization this week, according to The Information, which first reported the departures and said TechCrunch independently corroborated them. Neither outlet's brief named the researchers or stated an explicit reason, leaving the company's own account of what happened the only version in circulation so far.
A Safety Team With a History of Turnover
This isn't the first time OpenAI's safety function has turned over under pressure. The company's original Superalignment team, formed in 2023 to work on long-term AI risk, was dissolved in 2024 after co-leads Jan Leike and Ilya Sutskever both departed within weeks of each other, citing disagreements over how much priority safety work was getting relative to product shipping. OpenAI folded remaining safety staff into other research teams at the time. This week's firings suggest the underlying tension between a company racing to ship frontier models and a safety function meant to slow that process down when warranted hasn't gone away.
“- Total raised — approximately $180 billion since inception.”
The Regulatory Backdrop
The timing compounds the story. Axios reported Tuesday that the FTC has opened an inquiry into how both OpenAI and Anthropic handle their own safety claims — whether the labs' public statements about model risk match their internal evaluation processes. Separately, the UK's AI Safety Institute has flagged deceptive behavior in frontier agents from both companies, per Pulse's prior coverage. Regulators in two jurisdictions are now actively scrutinizing the exact function — pre-deployment safety evaluation — that these three researchers would have staffed.
Company Context
OpenAI, founded in 2015 and based in San Francisco, has roughly 900 million weekly active users and more than 50 million paid subscribers, per Pulse's company tracking:
- Revenue — roughly $25 billion annualized.
- Total raised — approximately $180 billion since inception.
- Last valuation — $852 billion, set in a March 2026 round backed by Amazon, Nvidia and SoftBank.
Its closest safety-focused peer, Anthropic, has built its entire public identity around safety research as a product differentiator — a positioning that makes any OpenAI safety-team shakeup an implicit comparison point whether or not Anthropic says anything about it.
Why Three Departures Is Hard to Read From the Outside
Outside observers have almost no way to calibrate what three firings actually mean inside an organization OpenAI's size. If the safety org numbers in the hundreds, as comparable functions at Anthropic and Google DeepMind reportedly do, three departures could be routine performance management with unfortunate timing. If the cut landed on a specific small team working a specific eval problem, it could be a meaningful signal about what OpenAI is deprioritizing. The Information and TechCrunch reports don't specify which, and OpenAI hasn't issued a public statement addressing the departures directly.
What the firings alone don't tell you: three departures from a large safety organization isn't necessarily evidence of a strategic deprioritization — people get fired for performance, policy violations, or reasons entirely unrelated to their specific function, and neither outlet has published documentation connecting these firings to any specific safety disagreement. The Leike/Sutskever exits in 2024 were voluntary and explicitly about strategy; this week's departures are involuntary and, so far, unexplained.
The detail worth tracking is whether any of the three researchers speaks publicly about why they left, the way Jan Leike did in 2024. Silence from people who were just fired is common and means nothing; a public account, if one comes, would settle whether this is routine personnel turnover or the same strategy fight resurfacing under a new set of names.