OpenAI eyed $5B+ acquisition of interpretability lab Goodfire
TL;DR
- OpenAI leaders including CEO Sam Altman held multiple summer talks about buying Anthropic-backed Goodfire for more than $5 billion.
- The proposed price was more than four times Goodfire's $1.25 billion valuation from its $150 million February Series B.
- No deal or partnership emerged; Goodfire's leadership preferred to remain independent and keep selling to multiple AI developers.
OpenAI explored paying more than $5 billion for Goodfire, an Anthropic-backed interpretability startup last valued at $1.25 billion, according to The Information.
Sam Altman and other OpenAI executives held multiple conversations with Goodfire over the summer, exploring either an outright purchase or a partnership. Neither came together. The startup's leadership "preferred to remain independent and sell its services to multiple AI developers," per the report, a notable stance given the proposed price was more than four times the $1.25 billion valuation Goodfire set in its $150 million February Series B.
The timing is the strange detail. The conversations happened as "OpenAI staff tried to understand how its AI caused security breaches inside and outside the company." Goodfire works on mechanistic interpretability, reverse-engineering a model's internal computations to see why it behaves the way it does. Its Ember platform counts the Arc Institute, Mayo Clinic, Microsoft, Baseten and Prima Mente among named partners.
Anthropic sits on Goodfire's cap table with a $1 million check that it described as its first direct investment in another startup. An OpenAI buyout would therefore have meant cashing out a rival investor and consolidating a safety tool used across the industry inside a single lab. Other investors include Menlo Ventures, B Capital, Lightspeed, Salesforce Ventures and Eric Schmidt.
OpenAI's appetite for interpretability work is not abstract this week. The same company just embedded Apollo, METR, Transluce and Goodfire in a safety audit, one of 406 OpenAI stories on our tracker in the last 90 days, and is simultaneously facing a California lawsuit from AG Bonta that cites 80 "loss of control" incidents tied to its own agents.
Neither company has commented publicly.
Originally reported by The Information
Read the original article →Original headline: OpenAI Explored $5B+ Acquisition of Anthropic-Backed Interpretability Lab Goodfire, No Deal Done