News

OpenAI weighed $5B deal for AI interpretability startup Goodfire

What's the deal? OpenAI explored acquiring Goodfire, a startup that analyses how AI models reach their outputs, for more than $5 billion, according to The Information . The talks did not produce a deal, and the final figure was never confirmed.

What does Goodfire do? The two-year-old, Anthropic-backed startup works on interpretability — reading the internal workings of AI models rather than just testing their outputs. That focus has gained relevance as labs confront model behaviour that is harder to monitor.

Why now? OpenAI has separately hired Goodfire, alongside Apollo ResearchDealroom has a profile for this one. Try Dealroom →, METR, and TransluceDealroom has a profile for this one. Try Dealroom →, for embedded safety assessments. Goodfire's role is to analyse model internals and monitor training runs, with OpenAI researchers the primary audience for its findings.

The price: The reported offer of more than $5 billion would mark a sharp jump from Goodfire's February Series B valuation of $1.25 billion — roughly four times higher. The startup walked away, betting it is worth more as an independent company.

What could go wrong? The paid arrangement is not an independent public audit, since OpenAI commissions and receives the work. That raises questions about how much scrutiny embedded safety partners can apply to the labs funding them.

The signal: The talks point to growing interest in interpretability as a complement to output-based testing. As AI systems grow more complex and their behaviour harder to track, the firms that can explain how models "think" are becoming strategic targets.

Read more: The Information

Image credit: "Quantum Computing for Google Goggles" by jurvetson, licensed under CC BY 2.0

More top stories