Anthropic taps Accenture's Faculty as its first embedded safety evaluator — $1B each over five years

An AI Frontier Post write-up, by Marcus Doyle, of Anthropic’s announcement on 18 September that Faculty, Accenture’s AI business, will be its first embedded evaluator. Faculty’s staff will work inside Anthropic with access comparable to an employee’s: watching models in training, following deployment decisions, talking to staff directly, and doing red-teaming, alignment assessments and safeguard testing. Each company expects to invest at least $1 billion over five years. The deal is non-exclusive, Anthropic is talking to METR and other nonprofits, and Anthropic pays for Accenture’s work for now while saying the long-term money should come from pooled or government sources. Doyle presents it as the first concrete step from Dario Amodei’s essay of 12 September, “We Must Pace the Frontier”.

His question is whether evaluation paid for by the company being evaluated can count as independent. The same day, more than 100 researchers, Geoffrey Hinton among them, signed a letter asking that embedded evaluators be “meaningfully independent”: not owned or governed by a frontier lab, with no other significant commercial business with one, and paid nothing that depends on their findings. Doyle’s reading is that Accenture passes the first test and strains the other two. He adds that the research community had expected nonprofits such as METR and Redwood Research to fill this role rather than a management consultancy, and closes on what to watch: whether inside access finds problems that outside testing misses, and whether the evaluators stay independent while being paid.

The terms match Anthropic’s own post and other coverage, and the letter’s three criteria match CNBC’s and implicator.ai’s reporting. One point the article leaves out sharpens its own argument: Accenture already has significant business with Anthropic, through the Accenture Anthropic Business Group formed in December 2025 to train about 30,000 Accenture staff on Claude and roll out Claude Code. Faculty itself only joined Accenture this year, announced on 6 January and completed on 16 March. Anthropic’s post says there is “no settled system” yet for funding independent evaluation, and the terms it gives on reporting say only that evaluators can report incidents and give the public a fuller account of risks. Nothing published so far says whether they can publish without Anthropic’s sign-off. The article links to none of its sources.