Mirendil raises $200 million to build self-improving AI that accelerates scientific discovery

Former Anthropic Researchers Launch Mirendil with $200 Million Seed Funding

Mirendil has secured $200 million in seed funding at a $1 billion valuation, marking one of the largest seed rounds in the AI sector1

1

. The startup was founded by Behnam Neyshabur and Harsh Mehta, former Anthropic researchers who left the company in January 2025 after spending barely over a year there1

1

2

2

. Andreessen Horowitz and Kleiner Perkins co-led the round, with Nvidia joining as an investor1

1

. The company currently operates with about 20 researchers and engineers drawn from Anthropic, xAI, Google DeepMind, and OpenAI1

1

.

Building AI That Improves AI for Scientific Discovery

Mirendil's mission centers on creating self-improving AI that does the work of an AI researcher—designing experiments, searching for optimal settings, evaluating models, and running subsequent training rounds1

1

. Neyshabur, who serves as chief executive and previously spent over five years at Alphabet co-leading reasoning research for Gemini, frames the platform as "AI for AI for science"1

1

. The goal is to enable organizations like university biology labs to build specialized models—such as drug-target prediction systems or Alzheimer's risk assessment tools—without requiring dedicated machine-learning teams1

1

. According to the founders, work that typically takes labs months could compress into days1

1

.

Breaking the Lock on Recursively Self-Improving AI Systems

The startup addresses a critical gap in AI research. Major AI labs use recursive self-improvement internally but prevent others from accessing this capability through restrictive terms of service1

1

. As of May, Anthropic reported that Claude wrote more than 80% of the company's own code, yet its terms forbid using the tools to build competing services1

1

. Anthropic defended this policy as standard among model providers, citing concerns about keeping frontier AI away from foreign adversaries1

1

. Matt Bornstein from Andreessen Horowitz told the Wall Street Journal that labs act as "rational economic actors" when denying customers the means to enhance their own models, adding that "structurally, there has to be an independent company"1

1

.

Democratize Access to AI Amid Industry Tensions

Mirendil's launch comes at a tense moment for AI development. Anthropic recently pulled access to its most powerful Mythos and Fable models after the Trump administration imposed export controls1

1

. The company also faced criticism for allegedly degrading answers about AI development1

1

. Both Anthropic and OpenAI have publicly called for global oversight committees to monitor recursively self-improving AI and enforce slowdowns if necessary to prevent loss of human control2

2

. The founders, however, view recursive self-improvement as the "shortest path" to faster science and believe it can be supervised rather than avoided1

1

.

Accelerate Scientific Discovery Through Open Access

The company's website states: "Today, any lab trying to use AI in drug development, chemistry, biology, or robotics must also become a frontier AI lab"2

2

. Mirendil aims to change this by making frontier AI research and development widely accessible2

2

. The startup believes the problem isn't self-improving AI itself but rather that access to such capabilities is currently gated by a small number of deep-pocketed AI labs2

2

. By extracting patterns from vast datasets and enabling agentic systems that improve over time, the platform could help independent laboratories push frontiers in their own domains of expertise2

2

.

AI Infrastructure Market Fuels Massive Valuations

The $200 million figure reflects broader trends in venture capital. AI captured close to half of all global venture funding in 2025, totaling approximately $202 billion—up more than 75% year-over-year according to Crunchbase1

1

. The AI infrastructure market alone ended 2025 near $337 billion in revenue and is forecast to reach $1.2 trillion by 20301

1

. Mirendil joins a cluster of high-profile lab spinouts: Ilya Sutskever's Safe Superintelligence raised $6 billion at a $32 billion valuation, while Mira Murati's Thinking Machines Lab secured $2 billion at $12 billion1

1

. The company's website features job postings with starting salaries of up to $500,0002

2

. The founding team includes Shayan Salehian, an early xAI member, and Tara Rezaei, a 23-year-old MIT graduate1

1

. Notably, Harsh Mehta built the first version of Anthropic's internal AI research platform, at times working as a team of one—now he's rebuilding that concept to sell it1

1

.