AI
Anthropic Just Invited the Auditor Inside the Lab, and Both Sides Put a Billion Dollars on the Table
On Friday, Anthropic announced it had selected Accenture as its first embedded evaluator, and the two companies each committed at least one billion dollars over the next five years to the effort. The work runs through Faculty, Accenture's specialist AI unit, and covers red teaming models, alignment assessments, and testing safeguards. The money is the headline. The access is the story.
Embedded evaluation puts outside auditors inside the building with access comparable to an employee's. They watch models take shape during training, follow the decisions that govern how those models are built and released, and speak directly with the staff making those calls. Anthropic frames the arrangement as a way to make its own accountability more verifiable. Until now, third party audits of AI systems happened at arm's length, against finished models, using the artifacts a lab chose to hand over. This inverts that. It is the difference between reading a company's report card and sitting in the classroom.
The move is the first concrete deliverable from chief executive Dario Amodei's September 12 essay, We Must Pace the Frontier, which called for frontier labs to open their doors to ongoing, employee like access for independent evaluators. The industry is already responding. OpenAI chief executive Sam Altman has said his company would give outside evaluators similar access, and Microsoft chief executive Satya Nadella welcomed embedded evaluators while cautioning that oversight should stay spread across many hands rather than sit with a few. The market liked the news too, and Accenture shares jumped as much as eight percent in after hours trading on the prospect of a new revenue stream in AI oversight.
The arrangement carries a live tension, and it surfaced the same day. More than one hundred researchers, including Geoffrey Hinton, signed a public letter calling for evaluators embedded in AI companies to be meaningfully independent. The concern is structural. Anthropic is paying Accenture directly for the work, at least in the near term, while hoping pooled industry or government funding sustains the model later. Anthropic says the arrangement is non exclusive in both directions and that more evaluators will be announced in the coming weeks. It is also talking with the nonprofit METR about pilots funded by METR itself, and METR says it accepts funding from neither AI companies nor their executives. Access solves one design challenge. Funding independence is the next one, and both facts sit inside the same announcement.
Faculty brings credentials that go beyond general consulting. The unit, which Accenture folded into its specialist AI business, has advised governments and defense organizations and built the United Kingdom National Health Service's COVID 19 early warning system. Its chief executive, Marc Warner, says the company was founded on the belief that AI should be safe by design rather than safe by accident. That real world perspective is exactly what Anthropic says it wants pointed at models headed for broad use.
For readers, the practical read is simple. Third party verification is becoming the price of selling frontier models into regulated industries, and the labs are starting to build that cost into their plans. Watch the next announcements, which evaluators come next, whether government money follows, and whether other labs match the access Anthropic just granted. The classroom door is open. The interesting part is who walks through it next.
Quick answers
What is this story about?
On Friday, Anthropic announced it had selected Accenture as its first embedded evaluator, and the two companies each committed at least one billion dollars over the next five years to the effort. The work runs through Faculty, Accenture's specialist AI unit, and covers red teaming models, alignment assessments, and testing safeguards. The money is the headline. The access is the story.
Why does this story matter?
For readers, the practical read is simple. Third party verification is becoming the price of selling frontier models into regulated industries, and the labs are starting to build that cost into their plans. Watch the next announcements, which evaluators come next, whether government money follows, and whether other labs match the access Anthropic just granted. The classroom door is open. The interesting part is who walks through it next.
Sources
- AI Weekly on the Anthropic Accenture deal
- Tech Xplore on the embedded evaluator announcement
- Edgen Tech on the two billion dollar evaluation commitment
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.