Andrej Karpathy announced on Monday that he has joined Anthropic, making him the latest — and arguably most prominent — member of what the industry is now calling the OpenAI exodus.
Of the eleven people who co-founded OpenAI in 2015, exactly two remain: CEO Sam Altman and President Greg Brockman. Everyone else has left. Karpathy’s move to Anthropic puts a sharp point on what has become one of the most significant talent migrations in tech history.
What Karpathy Will Do
Karpathy joins Anthropic’s pre-training team, the group responsible for the large-scale training runs that give Claude its core knowledge and capabilities. He’ll work under team lead Nick Joseph — himself a former OpenAI researcher — and will lead a new initiative focused on using Claude itself to accelerate pre-training research.
That last part matters. Anthropic is betting that AI-assisted research, rather than sheer compute alone, is how it stays competitive with OpenAI and Google. Having one of the field’s most respected researchers lead that effort sends a signal about where the company thinks the real breakthroughs will come from.
“I’ve joined Anthropic,” Karpathy posted on X. “I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D.”
He also noted that he “remains deeply passionate about education” and plans to resume that work in time — a reference to Eureka Labs, the AI education startup he founded in 2024 after leaving OpenAI for the second time. Eureka Labs hasn’t been formally shut down, but Karpathy’s full attention is now on pre-training at Anthropic.
The OpenAI Exodus by the Numbers
Karpathy is not an isolated case. At least 21 key leaders have left OpenAI since 2024, with 12 senior executives departing in 2025 alone. The list reads like a who’s-who of AI research:
- Ilya Sutskever, co-founder and chief scientist, left to start Safe Superintelligence (SSI)
- John Schulman, co-founder, went to Anthropic before becoming chief scientist at Thinking Machines Lab
- Mira Murati, CTO, left to found Thinking Machines Lab
- Bob McGrew, chief research officer
- Jerry Tworek, VP of research who led development of the o1 reasoning model
- Shengjia Zhao, key architect of ChatGPT and GPT-4, became chief scientist at Meta Superintelligence Labs
- Tim Brooks, who co-led Sora, went to Google DeepMind and then Meta
- Bill Peebles, Sora researcher, departed April 2026
- Kevin Weil, who led OpenAI for Science, departed April 2026
The departures accelerated in early 2026 after OpenAI signed a controversial AI collaboration contract with the U.S. Department of Defense. Some employees issued an open letter opposing the use of AI technology for mass surveillance or autonomous lethal weapons.
The safety-specific drain is particularly notable. By early 2026, essentially none of the people most associated with AI safety at OpenAI — the researchers who built alignment teams, the executives who advocated for caution, the board members who tried to enforce accountability — remained in positions of influence.
Where the Talent Went
The departures haven’t scattered randomly. They’ve clustered around a handful of destinations, each representing a different bet on AI’s future:
Anthropic has absorbed researchers who prioritize safety alongside capability. Schulman’s move there in 2024 was the early signal; Karpathy’s arrival confirms the trend.
Meta’s Superintelligence Labs pulled at least seven OpenAI researchers, including Zhao, drawn by Meta’s open-weight philosophy and massive compute resources.
Startups attracted the entrepreneurially minded: Sutskever’s SSI, Murati’s Thinking Machines Lab, and Fedus’s Periodic Labs each represent a different thesis about what AI development should look like.
The pattern tells a story. Researchers who care about safety and responsible development gravitate toward Anthropic. Those who want to work on open models go to Meta. Those who want to start fresh build their own labs. The one thing they have in common is that they all left OpenAI.
Anthropic’s Position
Karpathy arrives at a company on a steep trajectory. Anthropic’s annualized revenue hit $30 billion in April 2026, up from $9 billion at the end of 2025 — an 80x growth rate from January 2024’s $87 million run rate. The company is in early talks for a funding round at a valuation exceeding $900 billion.
Headcount has grown roughly 10x since early 2023, from 240 to an estimated 3,000–5,000 employees. Unlike many tech companies in 2026, Anthropic has conducted zero layoffs and currently has over 400 open positions.
The company is also fighting a legal battle with the U.S. government over its blacklisting — a consequence of the Mythos model’s ability to discover software vulnerabilities — while simultaneously serving as a magnet for researchers who believe AI development can be both commercially viable and responsibly managed.
What This Means
The AI talent war isn’t really about compensation packages, though those are certainly large. It’s about which vision of AI development researchers want to spend their most productive years on.
OpenAI’s pivot toward defense contracts, aggressive commercialization, and a potential IPO at a $157 billion valuation has clarified what the company prioritizes. That clarity has made it easier for researchers to decide whether they want to be there.
Anthropic’s pitch — safety research as a competitive advantage, not a compliance cost — is clearly winning converts. Karpathy doesn’t need the money or the resume line. He chose to work on pre-training at a company that publicly commits to responsible development, rather than return to the one he helped found.
For the broader AI industry, the concentration of elite talent at Anthropic, Meta, and a handful of startups means the next wave of breakthroughs won’t come exclusively from OpenAI. The company that once had an unmatched monopoly on AI research talent now faces serious competition from labs staffed, in many cases, by its own alumni.
The question OpenAI should be asking isn’t how to stop the bleeding. It’s why the people who know the company best keep choosing to leave.