A specific type of institutional signal is absent from press releases and earnings reports. It appears in quiet job announcement posts, LinkedIn notifications, and the gradual accumulation of names—known and significant names—appearing on the departing side of a company’s ledger. For some time now, OpenAI has been generating that signal. Ignoring it is becoming more difficult.
Waves of departures have been coming in. The company’s chief technology officer, Mira Murati, who had been there long enough to be regarded as part of its foundational architecture, abruptly left. Barret Zoph, the vice president of research, and Bob McCrew, the chief research officer, also did. Next, another vice president of research, Jerry Tworek.
Key Facts: The OpenAI Talent Exodus
| Company | OpenAI — maker of ChatGPT; valued at ~$300 billion; headquartered in San Francisco, CA |
| CEO | Sam Altman — has publicly prioritized rapid ChatGPT improvements over long-term research (“code red” directive) |
| Key departures | Mira Murati (CTO), Bob McCrew (Chief Research Officer), Barret Zoph (VP Research), Jerry Tworek (VP Research), Tim Brooks (Sora lead → Google DeepMind), Tom Cunningham (economist), Andrea Vallone (model policy) |
| Research job share shift | General research roles fell from 23% of job postings (2021) to just 4.4% (2024) — per Lightcast data |
| Stated reason for shift | Pressure from Microsoft partnership and commercial targets; prioritizing ChatGPT reliability over experimental research |
| Internal tension reported | Tom Cunningham alleged OpenAI was reluctant to publish studies that cast AI negatively; leadership remained silent |
| Key rivals benefiting | Google DeepMind, Anthropic (Claude series), Meta AI — all actively recruiting OpenAI alumni |
| Parallel at Meta | Yann LeCun departed citing manipulated Llama 4 benchmarks, internal conflict, and mass resignations post-launch |
| Investor response | $6.6 billion raised despite ongoing departures — investors largely unbothered, for now |
| Reference | OpenAI Official Website |
Then, Tim Brooks, who had been in charge of one of OpenAI’s most well-known recent bets, the Sora video generation project, declared he was joining Google DeepMind. The research team’s economist, Tom Cunningham, reportedly sent an internal email before departing, claiming that the company had become hesitant to publish studies that might present AI in a complex light. There was no public response from the leadership. In that situation, silence usually conveys something.
A version of the same frustration is shared by the departing researchers, at least according to what a number of them and those close to them have described. They claim that OpenAI has changed from being a company that made bold research bets that weren’t always related to a product timeline to one that is focused on improving ChatGPT’s speed, competitiveness, and appeal to Microsoft and the enterprise market. An anonymous former employee told WIRED, “People who like to do research are being forced to do product,” out of concern for their professional relationships. Even in its most basic form, that sentence probably captures the main conflict better than any organizational chart.
The figures are pointed. Twenty-three percent of OpenAI’s job postings in 2021 were for general research positions, according to data gathered by Lightcast. That percentage fell to 4.5 percent by 2024. It’s not a drift. Even though it wasn’t declared as such, that is a purposeful reorientation. The company has been discreetly reorganizing itself into something that looks much more like a software product company with excellent public relations, after initially positioning itself as a research organization working toward artificial general intelligence.
It’s important to consider what was going on when this change picked up speed. When ChatGPT first appeared as a research preview in late 2022, it quickly became a cultural phenomenon—a time when something that had previously primarily been discussed in the industry suddenly became accessible to everyone. It is easy to imagine the pressure that followed that achievement. Microsoft needed dependability and improvement on a commercial schedule because it had heavily relied on the OpenAI partnership for its cloud and productivity goals.
Rivals weren’t motionless. Gemini was incorporated into Google’s enterprise suite and Search. Anthropic, founded in 2021 largely by people who had themselves left OpenAI over concerns about safety and research direction, positioned Claude as the more research-serious alternative and kept attracting talent and capital. The race tightened faster than most observers expected, and the natural response to a tightening race is to run faster — which, organizationally, tends to mean cutting what doesn’t directly contribute to speed.
The leadership of OpenAI would characterize all of this in a different way, and it’s not wholly incorrect. Technically significant work has continued to be produced by the company. It recently raised 6.6 billion dollars, and investors didn’t seem to care too much about the internal unrest.
The model releases have continued to follow the stated pillars of safety, performance, and utility. The brain-drain narrative is refuted by the argument that every mature business eventually leans toward product, and that a business that can raise that much money at that valuation is not clearly in decline. It’s possible the departures represent growing pains rather than structural damage.
And yet. Watching this unfold over the past year and a half, there’s a feeling that something genuinely consequential is being lost — not just individual researchers, but a certain institutional disposition, the willingness to pursue questions that don’t connect cleanly to a shipping date. The departing employees are not quietly retiring at a young age.
They will work at Anthropic, DeepMind, academic institutions, and their own startups. They are carrying their inquiries with them. And the fields those researchers work in — alignment, interpretability, novel architectures — are exactly the areas where OpenAI built its early reputation and where the long-term competition in AI will likely be decided.
It’s still unclear whether the next generation of researchers entering the field will continue to view OpenAI as the obvious destination it once seemed, or whether the accumulation of departures will slowly shift that perception. For now, the talent keeps moving. And the companies receiving it are paying very close attention.
