Newshot

OpenAI and Goodfire strike AI safety partnership after earlier talks over a $5 billion-plus acquisition

AI summary

OpenAI announced partnerships with Apollo Research, METR, Transluce and Goodfire to launch the first phase of its Embedded Assessments program, working together on model interpretability and safety evaluation. The Information reported that OpenAI had previously explored acquiring Goodfire for more than $5 billion, but the two sides ultimately failed to reach a deal. Under the partnership, Goodfire will send interpretability researchers into OpenAI to monitor training using its own analysis platform and gain deep access to internal models and systems, with a focus on model forensics.

Translation(model translation, not the source's own text)

OpenAI and Goodfire Reach AI Safety Partnership After Previously Discussing an Acquisition Worth Over $5 Billion

IT之家 November 17 news: Today, OpenAI announced partnerships with Apollo Research, METR, Transluce, and Goodfire to launch the first phase of its Embedded Assessments program, jointly conducting work on model interpretability and safety evaluation.

Meanwhile, according to a report by The Information today, OpenAI had previously explored a plan to acquire Goodfire for more than $5 billion (IT之家 note: approximately 33.392 billion RMB at current exchange rates), but the two sides ultimately failed to reach an acquisition agreement.

According to an announcement published by Goodfire, the company will send interpretability researchers to work inside OpenAI, using its own interpretability analysis platform to monitor the model training process, investigate unexpected behaviors, and help improve model alignment and safety. The researchers will receive deep access to OpenAI's internal models and systems, specifically including internal signals during training and related technical resources.

The focus of this partnership is Model Forensics. Goodfire will analyze the internal computational processes of models to monitor known risks and anomalous behaviors that arise during training, helping OpenAI researchers discover issues that traditional monitoring methods may miss. The evaluation results will be provided primarily to OpenAI's internal research team to improve model safety. Goodfire also stated that, as the partnership progresses, it will gradually publish some of the research results.

In its announcement, OpenAI stated that embedded assessments allow independent experts to engage deeply with the lab's internal models and systems over an extended period, evaluating model alignment and the effectiveness of safety safeguards. Depending on the needs of different assessment projects, external evaluators may access model checkpoints, source code, training and evaluation data, monitoring systems, technical documentation, as well as the researchers and engineers responsible for the relevant work.

On the other hand, according to The Information, OpenAI approached Goodfire multiple times this summer, and OpenAI CEO 萨姆 · 奥尔特曼 also participated in the related discussions. At the time, OpenAI was investigating internal and external safety incidents caused by AI models and hoped to use Goodfire's technology to gain a deeper understanding of model behavior.

People familiar with the matter said that a senior OpenAI executive had proposed acquiring Goodfire for more than $5 billion, or pursuing other forms of commercial partnership. Based on the acquisition offer, the valuation would be more than four times the $1.25 billion valuation Goodfire announced during its fundraising in February this year.

However, the two sides ultimately did not reach an acquisition agreement. Goodfire's management believed that remaining independent and selling its technology to multiple AI developers could bring higher long-term value to the company.

The Information also mentioned that Anthropic invested $1 million in Goodfire in 2025 (approximately 6.678 million RMB at current exchange rates), marking Anthropic's first direct investment in another startup. Anthropic CEO 达里奥 · 阿莫迪 said at the time that the investment reflected the company's emphasis on mechanistic interpretability technology and its hope to transform black-box neural networks into systems that are easier to understand and control.

Related Articles

  • Anthropic CEO 阿莫迪: Compute Is the Deepest Business Challenge
  • Without Calling an Image Model, Anthropic Claude Fable 5.2 Uses Python Code to Simulate Real Brushstrokes
  • Reserve Bank of Australia Study: AI Hits Young Workers First, Especially in Software and Customer Service Roles
  • Former OpenAI Researcher: People Alive Today Are Most Likely to Die from AI Going Out of Control
  • Developer Uses Claude to Build 97KB Underwater Exploration Simulation Project Abyssal Garden
  • Former Anthropic Researcher Coxon: Retraining Mythos Today Would Require Only a Small Fraction of the Initial Compute

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗