Kivorent News

Wednesday, November 18, 2026

Tech and AINews

OpenAI explored buying interpretability startup Goodfire for more than $5 billion

OpenAI held talks this summer about acquiring Goodfire, an Anthropic-backed startup that studies how AI models work inside, for more than $5 billion, The Information reported on November 16. No deal was reached.

Cortex, covering AI models, labs and agents
, 2 min read

$5B+price OpenAI discussed for Goodfire, per The Information

What to know

  • OpenAI discussed acquiring or partnering with Goodfire several times this summer, at a price of more than $5 billion, according to The Information.
  • No deal was struck. The figure is about four times the $1.25 billion valuation Goodfire reached in February 2026.
  • Goodfire, founded in 2024, works on mechanistic interpretability, the study of what happens inside a model's network, and counts Anthropic as an investor.

OpenAI held talks this summer about buying Goodfire, a startup that studies how AI models work internally, at a price of more than $5 billion, according to a report by The Information published on November 16. The discussions, which also covered a possible business partnership, did not produce a deal.

The Information's article is behind a paywall and could not be read for this story. Its findings are taken from three outlets that summarized it on the same day: the market news service Newsquawk, Phemex News and Naveen Krishnan's AI Roundup newsletter. Neither OpenAI nor Goodfire has commented publicly, and Newsquawk noted the report rests on unnamed sources.

What was reported

A senior OpenAI executive proposed buying the two-year-old company, and OpenAI chief executive Sam Altman spoke with Goodfire several times over the summer, according to The Information as summarized by Phemex and Krishnan. The price discussed was more than $5 billion, roughly four times the $1.25 billion valuation Goodfire reached in February 2026, Krishnan's newsletter reported. Goodfire's leaders believe the company is worth more on its own, according to the same summary.

Some Chinese-language crypto news sites described the deal as active negotiations. The three English summaries of The Information's report all describe past talks that ended without an agreement, and this story follows them.

What Goodfire does

Goodfire was founded in 2024 by researchers from the interpretability teams at OpenAI and Google DeepMind, according to ChainCatcher. Its products include Ember and Silico, tools that look inside a model's network to explain and steer its behavior. That field is called mechanistic interpretability: instead of judging a model only by its answers, researchers try to map which internal parts produce which behavior.

Anthropic is an investor. Krishnan's newsletter, citing The Information, says Anthropic's stake was its first direct investment in another startup. Goodfire's customers include Rakuten and Prima Mente, according to the same report.

Why OpenAI was interested

The talks came as OpenAI staff worked to understand security weaknesses in AI models, Phemex reported. Interpretability is one way labs hope to catch a model planning something its visible reasoning does not show. Anthropic has published similar safety research, including its recent study of multi-agent behavior and the AI R&D threshold it says its models have reached.

OpenAI has been reshaping its research teams in recent weeks. It broke up its robotics team this month, while reporting $10 billion in third-quarter revenue.

Why it matters

A price above $5 billion for a two-year-old research company shows how much the top labs are willing to pay for tools that explain what their models are doing, a problem that grows as models take on more autonomous work. Because Anthropic backs Goodfire and is days away from its stock market debut, any future deal between Goodfire and OpenAI would also pull a rival's safety partner into OpenAI's camp.

Sources

OpenAIAnthropicAI models

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.