AI Policy Wiki
Dashboard

Mirendil

low confidence · updated 2026-08-06

AI startup founded by Anthropic veterans, working on self-improving AI systems with the stated aim of eventually taking on the work of an entire frontier AI lab. Raised seed funding at a $1 billion valuation in late June 2026 and signed a multi-year Google Cloud compute agreement worth more than $100 million, disclosed August 6, 2026.

Mirendil is an AI startup founded by veterans of Anthropic and led by co-founder and chief executive Benham Neyshabur. The company states it aims to build systems that can eventually take on the work of an entire frontier AI lab, and describes its research programme as self-improving AI (Source: techcrunch.com).

Funding and compute

Mirendil raised seed funding at a $1 billion valuation in late June 2026. On August 6, 2026 it was reported to have signed a multi-year partnership with Google Cloud worth more than $100 million to source compute for that research. Neyshabur described the commitment as roughly half of what the company raised in the seed round, which places the round at approximately $200 million on that arithmetic — a figure implied by the ratio rather than separately disclosed (Source: techcrunch.com).

Committing about half of a seed round to a single cloud contract is a compute-to-capital ratio at the high end of the pattern tracked at AI Infrastructure Capex, and the arrangement places a company pursuing self-improving systems on a hyperscaler's infrastructure rather than its own.

Research position

The stated objective — systems capable of doing the work of a frontier lab — is the strongest form of the claim examined at Recursive Self-Improvement (RSI), and places Mirendil alongside Discovery Loop, founded a day earlier by departing Google researchers on a comparable premise. Neither company has published technical results. What such systems would need to do beyond verifiable engineering tasks is the subject of Can AI agents conduct open-ended AI research? Early evidence from two case studies (Kirgis et al., July 2026) and Verification Asymmetry.

The record on this company rests on a single report, and the founding date, headcount, seed-round total and investor list are not established here.

Relationships