Who is Andrej Karpathy: from OpenAI through Tesla back to the core of AI

Thirty-nine-year-old Slovak-Canadian researcher Andrej Karpathy belongs to the innermost elite of global artificial intelligence. In 2015, he stood among the 11 original founders at the birth of OpenAI — back then still a nonprofit organization, from which grew the company worth hundreds of billions of dollars behind ChatGPT.

After leaving OpenAI in 2017, he headed to Tesla, where as director of AI and computer vision he led the development of systems for Autopilot. Under his leadership, Tesla built its own infrastructure for training neural networks, its own inference chip, and a complete pipeline from data labeling to deploying models in real vehicles.

In 2023, he returned to OpenAI — this time with a mission to build a team focused on midtraining and synthetic data generation. It was work directly following up on the pre-training of large language models, which is exactly the area he will focus on at Anthropic. He left OpenAI in 2024 and subsequently founded Eureka Labs, a project focused on AI-powered education, and created several open-source tools including the popular repository autoresearch and the LLM101n course.

What Karpathy will do at Anthropic

Karpathy joins the pretraining team, led by Nicholas Joseph — also a former OpenAI researcher who serves as Head of Pretraining at Anthropic. According to the company's official statement, Karpathy's team will use Claude to accelerate the pre-training research itself.

This is a significant shift toward recursive self-improvement of AI: the model helps researchers develop a better version of itself. This approach is referred to in expert circles as "AI-assisted AI research" and represents one of the key trends of 2026.

Karpathy's experience with midtraining and synthetic data from OpenAI is exceptionally relevant for this role. Synthetic data — that is, training data generated by another AI model — is becoming an increasingly important part of developing new generations of language models, especially at a time when high-quality human-written texts on the internet are running out.

Why he chose Anthropic

In his announcement on X (formerly Twitter) on May 19, Karpathy wrote: "I think the next few years at the frontier of LLMs will be especially formative. I'm excited to join the team and return to research and development." He also added that education remains his deep passion and he plans to return to it over time.

His choice of Anthropic is significant. At a time when OpenAI, Google, Meta, and Elon Musk's xAI are aggressively recruiting top talent, one of the most respected AI researchers chose precisely the company that has long built its identity on safety and responsible development. It's a signal that Anthropic can attract the elite even in a hypercompetitive environment.

Anthropic on the rise: SpaceX, Google, Amazon

Karpathy's hiring comes at a time when Anthropic is massively expanding. Just in May 2026, the company announced:

• A partnership with SpaceX, which will deliver over 300 megawatts of computing capacity within a month — enabling an immediate increase in usage limits for Claude users.
• A $40 billion investment from Alphabet (Google), which secured Anthropic access to additional computing power.
• An expansion of the alliance with Amazon, which had already invested billions into Anthropic.
• A collaboration with Broadcom and Google to secure 3.5 GW of next-generation computing capacity.

Anthropic also recently introduced Claude Opus 4.7, which according to benchmarks competes with the latest models from OpenAI, and launched specialized agents for finance, law, and creative tools. The company, according to available information, plans revenue exceeding $10 billion in 2026.

What this means for Czechia and Europe

For Czech users and businesses, it is significant that Claude is fully available in Czech — it understands it, responds in it, and the quality of Czech output in the latest models (Sonnet 4, Opus 4.7) is at a very high level. Every acceleration of Claude's development thus directly translates into tools that a Czech student, developer, or entrepreneur can also use.

In the European context, Karpathy's move is also interesting because he is a researcher with Slovak roots, who studied at the University of Toronto and the University of British Columbia — thus outside the traditional American axis. His career path shows that even researchers from Central Europe can stand at the absolute top of global AI.

From the perspective of the EU AI Act, it is important that Anthropic is among the companies that most vocally support safety regulation. Karpathy's focus on using Claude in research may also bring new tools for model auditing and interpretability, an area that European regulation emphasizes.

Comparison: Anthropic vs. OpenAI — the talent war

The move of such a prominent figure illustrates how the dynamics between the two companies have changed. Just two years ago, leaving OpenAI for Anthropic would have been seen as a step back. Today, when Claude has surpassed OpenAI in both revenue and enterprise customers for the first time, the situation is reversed.

Karpathy is not the first former OpenAI researcher to head to Anthropic — he joins colleagues like Nicholas Joseph, Jan Leike (former head of OpenAI's safety team), and John Schulman (another OpenAI co-founder). This concentration of talent gives Anthropic a unique advantage in the race for the next generation of language models.