Anthropic names Accenture as its first embedded evaluator, with employee-level access

Every high-risk industry eventually invents the same job. Someone is allowed inside the building, sees the work while it is being done, and is not paid to make it look good. Frontier AI has argued for two years about whether that job should exist. Anthropic has now named the first firm to do it.[1]

Accenture will build a team of embedded evaluators working alongside Anthropic's own safety teams. The published scope is evaluating and red-teaming models, conducting alignment assessments and testing model safeguards. Both companies say they each expect to invest at least one billion US dollars over five years.[1][2]

The partnership is led by Faculty, Accenture's specialist AI business. Accenture says Faculty already tests and evaluates models for other leading AI labs, and that its earlier work spans government, defence, healthcare and infrastructure, including the National Health Service's early warning system during the COVID-19 pandemic.[2]

Julie Sweet, Accenture's chair and chief executive, said safety requires both deep technical expertise and a clear understanding of how AI is used in the real world. Marc Warner, its chief technology officer and Faculty's chief executive, said AI should be safe by design, not safe by accident.[2]

Today's external evaluators test a finished model from outside it. Anthropic says embedded evaluators will work inside the company with access comparable to an employee's, watching models take shape during training, following the decisions that govern how they are built and deployed, and speaking directly to employees.[1]

Anthropic's chief executive set out the mechanics in an essay carrying a September 2026 dateline on his own site. It describes desks in Anthropic's offices, access badges, company laptops, and permissions mostly comparable to those of the internal teams that already assess risk.[3]

The essay says reviewers should have the right to publish key findings about risk levels, incidents, practices and the access they did or did not receive, without editorial control by Anthropic. It reserves a narrow right to redact security-sensitive, legally privileged or commercially sensitive material.[3]

Findings cannot be redacted merely for being unfavourable, and reviewers may say publicly when a redaction removed something important to their conclusions. All of this describes a contract Anthropic intends to sign, not a published agreement. Model safety, it says, remains its own responsibility.[3][1]

There are no standards yet for what information embedded evaluators should see, or how they should report what they find. Anthropic says so plainly. It adds that no settled system exists for funding independent evaluation, and that the money should eventually come from pooled or government sources.[1]

Because neither exists, Anthropic will fund Accenture's work directly. An evaluator paid by the party it evaluates is familiar from financial audit, and a familiar source of argument. Anthropic says it is also in dialogue with METR and other non-profit evaluators.[1]