Tech News
Anthropic and OpenAI propose embedding independent safety evaluators within AI companies
Anthropic and OpenAI have proposed giving third-party evaluators embedded access to assess AI safety, but questions remain about whether oversight will truly be independent.
Ceeve Team · 2026-09-16 · 2 min
A neutral summary of Anthropic and OpenAI want to embed safety evaluators. Will they really be independent?, published by TechCrunch. Summarised for the Ceeve reading list; the original reporting is TechCrunch's.
Anthropic CEO Dario Amodei and OpenAI's Sam Altman have proposed embedding independent third-party evaluators inside frontier AI companies to assess safety practices and report findings publicly. Organizations like METR and Redwood Research would gain unprecedented access to company systems to evaluate whether AI models are genuinely aligned. While external evaluators welcomed the proposal, they flagged concerns about whether the arrangement would guarantee true independence rather than simply creating vendor relationships.
Third-party evaluators want access beyond final models to intermediate checkpoints created during training, interview transcripts, and employee interviews to verify safety claims. This deeper access matters because models can be trained to perform well on safety tests whilst concealing problematic behaviour. Previous evaluations have faced constraints: METR and Redwood received only a week to investigate a Hugging Face incident, whilst Apollo Research had just three days to assess GPT-6 Astra, both limiting their ability to draw confident conclusions.
Whether Anthropic and OpenAI will actually grant the access and freedom evaluators need remains unclear, as neither company has detailed which evaluators will participate, timelines, or what information they can disclose. Researchers called for legally mandated frameworks rather than voluntary commitments, noting that companies controlling their own safety oversight whilst asking for public trust represents a contradiction; those job-hunting in AI safety should monitor how employers respond to these proposals via resources like Ceeve when assessing workplace credibility.
Read the original on TechCrunch
Ceeve tailors your CV to a specific role, writes the cover letter that goes with it, and prepares you for the interview. If a story here is about a company you would like to work for, Ceeve can research it, match your experience against the posting, and rehearse the questions with you before you walk in.