Anthropic Alumni Launch Biology-Focused AI Startup With $1 Billion Valuation
Behnam Neyshabur and Harsh Mehta, who left Anthropic in December 2025, have founded Mirendil, an AI startup targeting scientific breakthroughs in biology and materials science. The San Francisco company is in funding negotiations for $175 million at a $1 billion valuation, with Andreessen Horowitz and Kleiner Perkins co-leading discussions.
Neyshabur, the CEO, previously led Anthropic's scientific AI reasoning team and spent more than five years at Google DeepMind. Mehta, the CTO, was a senior research scientist at Anthropic. The founding team also includes Shayan Salehian from xAI and Tara Rezaei, a former OpenAI intern.
Why This Matters for Research Teams
Mirendil represents a shift in how frontier AI expertise gets applied to specific scientific domains. Rather than building general-purpose models, the startup plans to develop specialized systems for hypothesis generation and simulation in biology and materials discovery.
The $1 billion valuation in early funding talks signals investor confidence that domain-specific AI can outperform general models in hard sciences. This contrasts with the emphasis on broad capabilities that dominated AI investment in recent years.
Part of a Larger Trend
Mirendil joins a wave of specialized startups founded by researchers leaving major AI labs. These "neo-labs" typically focus on narrow applications-scientific research, office productivity, or new AI development methods-rather than competing in the general-purpose model space.
The movement reflects talent mobility from safety-focused labs like Anthropic to ventures that apply frontier AI to physical sciences. Backing from deep tech investors like Andreessen Horowitz validates the strategy of targeting niche applications where specialized expertise matters.
What Comes Next
If the funding round closes, Mirendil could scale its team and compute resources to train models for drug discovery and advanced materials research. The startup will face competition from established players like Google DeepMind, which already has resources and institutional knowledge in scientific AI.
Key milestones within the next 12 to 18 months include model releases and early research partnerships. The broader neo-labs movement may accelerate as geopolitical pressures continue to reshape where AI research happens and who leads it.
Your membership also unlocks:
