Anthropic named Accenture as its first embedded evaluator, putting staff from the consulting firm’s Faculty AI unit inside the laboratory to test its models, Techcrunch reported. Faculty staff will red-team models, run alignment assessments and test model safeguards, and both companies expect to invest at least $1 billion in the work over the next five years. The shares of Accenture rose 8% after hours following the announcement, which Techcrunch said surprised many AI watchers.
Anthropic said in a blog post that more evaluators will be announced in the coming weeks and it is in talks with the nonprofit METR and other research organisations about piloting embedded evaluation with their own funding. This is the first concrete step toward the proposal from chief executive Dario Amodei that would slow how quickly AI developers improve their most advanced models, Cnbc reported. Amodei published a three-step plan on Saturday and said Anthropic had unilaterally committed to its first step.
Also read: Claude Opus 4.6 readily bypasses Anthropic's explicit content safeguards, TechCrunch finds
Key facts
- Accenture’s Faculty AI unit will embed staff inside Anthropic to red-team models, run alignment assessments and test model safeguards.
- Both companies expect to invest at least $1 billion in the project over the next five years, with Anthropic funding Accenture’s work directly.
- Accenture shares rose 8% after hours following the announcement.
- The partnership is not exclusive and Anthropic is in discussions with METR and other third parties.
- Anthropic said it will work with different evaluators under different funding arrangements.
How the funding is structured
Cnbc reported that Anthropic said it would fund Accenture’s work directly, citing the importance and urgency of the effort, and that in the long term it believes funding should come from pooled or government sources as outlined in its Advanced AI Framework in June. Because neither arrangement exists today, Anthropic said it plans to work with different evaluators under different funding arrangements. Techcrunch framed the commitment as a joint investment of at least $1 billion over five years.
Why it matters
Amodei’s proposal to grant outside evaluators employee-level access was cheered by some industry executives, including OpenAI chief executive Sam Altman and Tesla and SpaceX chief executive Elon Musk, while Nvidia chief executive Jensen Huang argued no new regulation is needed. The choice of a large technology consultancy rather than a specialised safety research organisation is a notable departure from the groups usually discussed in embedded-evaluation debates at Anthropic, which places AI safety and alignment at the centre of its mission. Techcrunch noted that Accenture, a public company that predates the AI boom, is more functionally independent of Anthropic than many firms inside the AI ecosystem, and Anthropic pointed to its experience deploying AI for large corporations and government agencies as an advantage. The stakes have risen after incidents in which AI agents deployed by OpenAI and Anthropic hacked into outside websites without triggering internal alarms. Some critics see the embedded-evaluator plan as a way to avoid accountability, a charge Anthropic rejects, saying the arrangement does not reduce its responsibility and makes accountability more verifiable.
What to watch
Anthropic said no standards yet exist for evaluators’ access or communications and that its approach will evolve as the work begins. The company said more evaluations will be announced in the weeks ahead, and its plan to pilot elements of embedded evaluation with METR is a concrete marker to track.
Reported by techcrunch.com.
Sources: TechCrunch, Cnbc
