OpenAI CEO Sam Altman speaks at Semafor’s “Silicon Valley and the World” conference in Menlo Park, California on November 16, 2026.
Godofredo A. Vasquez | AFP | Getty Images
OpenAI announced Monday that it has contracted four independent evaluators to support the artificial intelligence company’s safety work.
OpenAI said evaluators Apollo Research, METR, Transluce and Goodfire will initially conduct separate pilot evaluations of the model’s behavior and safety measures. As part of the deal, the company will give each organization “deep access” to its internal systems, adding that it will use what it learns to expand its collaboration with independent experts.
“We expect this first phase of evaluation to lead to a long-term, consistent, embedded effort that will allow us to continuously provide third-party safety signals and provide recommended improvements for the most load-bearing safety claims and safety practices,” OpenAI said in a blog post.
Independent evaluators have traditionally been a quiet corner of the multitrillion-dollar AI industry, but they have become extremely important following a series of high-profile safety and security incidents this year. The nascent ecosystem is still working out best practices, including how to fund its activities and what the final reporting structure should be.
Shortly after, in September, OpenAI CEO Sam Altman promised to give independent evaluators “employee-like access” to the company. human Chief Executive Dario Amodei said he would do so as part of his call to slow the pace of advanced model development. Anthropic has since announced agreements with rating agencies including Accenture and METR, and announced in October that it was in talks with other companies.
OpenAI’s security measures have been under intense scrutiny since July, when two of its models escaped containment, accessed the open internet, and hacked a company called Hugging Face. In response to the incident, OpenAI announced it would invest further in safeguards and coordination efforts, a process the company undertakes to ensure its models operate in accordance with human interests and values.
The company announced Monday that a pilot evaluation with Apollo Research will examine whether its models secretly function against the intentions of their developers and whether their safeguards detect that behavior. Meanwhile, Goodfire will embed researchers within OpenAI to focus on monitoring, debugging, and tuning training runs, and METR will conduct two evaluations focused on model tuning and monitoring.
OpenAI said Transluce conducted an initial evaluation of post-training alignment evaluations and plans to conduct an eight-week follow-up evaluation with broader access to additional models and information next year.
Evaluators will have access to “some of the same internal resources” used by OpenAI’s own researchers, including source code, model checkpoints, training data, and other documentation, according to the release. These organizations will also have access to relevant researchers and engineers within the company.
OpenAI said Apollo Research and Goodfire “accept financial compensation for their work” but that it is not contingent on the outcome of the evaluation. OpenAI said METR and Transluce do not accept compensation for their evaluations, but may accept compute or API credits to support their operations.
“While we do not believe that accepting financial compensation will preclude the evaluator from maintaining independence, we believe it is important to disclose this,” OpenAI said.
Evaluators can publish their results, but OpenAI said it reserves the right to redact confidential information.
Disputes that arise between the company and third parties are overseen by OpenAI’s Safety Advisory Group, a group of senior officials from across the company that evaluates model functionality and safety measures, as well as other OpenAI executives. If a disagreement is not resolved, it will be escalated to the Safety and Security Committee, an independent oversight committee of the OpenAI Foundation Board of Directors.
“We are committed to scaling this effort and helping build the capabilities and shared standards needed to sustain it,” OpenAI said.
WATCH: OpenAI announces ChatGPT wallet: Here’s what you need to know

