Eight researchers walked away from OpenAI with Dario Amodei in 2021, collectively having shaped GPT-3, the neural scaling laws that define modern AI development, and the interpretability techniques that the safety field now depends on. Five years on, the pattern has begun repeating inside Anthropic itself: two teams of former Anthropic researchers have raised a combined $680 million in seed funding for their own AI labs in 2026. All eight of Anthropic's original co-founders carry a StartupHub.ai total score of 92 or above, a concentration of research credibility that rivals any single lab in the industry.
The Eight Who Left OpenAI
Dario Amodei departed OpenAI in December 2020, leaving behind a VP of Research role he held while co-leading the team that produced GPT-3, one of the most cited language model papers in AI history. Seven colleagues joined him in the months that followed. Daniela Amodei had served as VP of Safety and Policy at OpenAI. Tom Brown was the lead author of the GPT-3 paper, which demonstrated for the first time that scaling compute and data could produce emergent capabilities at a qualitative level. Jared Kaplan co-authored the neural scaling laws paper that established the mathematical relationship between model size, dataset size, and performance, a framework every major AI lab now uses when planning training runs. Christopher Olah had built the interpretability research discipline across Google Brain and OpenAI. Sam McCandlish, Jack Clark, and Ben Mann covered architecture, policy, and infrastructure, respectively, according to Penchan's founding history.
The stated reason for leaving was disagreement with OpenAI's approach to AI safety governance after it formed its commercial subsidiary in 2019. Anthropic's Long-Term Benefit Trust structure was designed to place safety objectives above investor returns. The company raised $300 million in its first funding round and has since scaled to above $9 billion in annual revenue, per Bloomberg. All eight original founders appear in the StartupHub.ai people database: six carry a total score of 97 or above, with Tom Brown topping the cohort at 99.
In August 2024, a notable figure extended the network further. John Schulman, one of OpenAI's six original co-founders and the researcher whose work on reinforcement learning from human feedback became central to ChatGPT's architecture, announced he was joining Anthropic to focus on hands-on alignment research. His tenure was brief. By February 2025, he had left for Mira Murati's Thinking Machines Lab, Winbuzzer reported, following the pattern of serial moves between frontier labs that has become a structural feature of the AI talent market.
Anthropic's Own Diaspora Takes Shape
The spinout cycle that made OpenAI the defining talent incubator of the early 2020s has begun inside Anthropic. In late 2025, Behnam Neyshabur and Harsh Mehta left the company, the former having spent seven years on AI for science across Google and Anthropic, the latter having built Anthropic's internal autoresearch platform from scratch. They co-founded Mirendil alongside Shayan Salehian, a machine learning veteran from xAI, and Tara Rezaei, a 23-year-old MIT graduate. The company closed a $200 million seed round at a $1 billion valuation from Andreessen Horowitz, Kleiner Perkins, and NVIDIA, TechFundingNews reported. Its focus is recursive self-improvement for scientific discovery, targeting biology and materials science applications.