On Friday the 18th, Anthropic appointed its watchdog — and, in the same press release, wrote the check. The Accenture deal is the moment "independent evaluation" stopped being a paragraph in an essay and became a five-year, billion-dollar-a-side commercial contract, signed six days after the essay, on the same day the world learned the company is pacing past $100 billion in revenue toward a November IPO. The loaded part is not the size of the commitment; it is the direction of the money. Anthropic is funding the evaluator directly because the pooled, government funding its own June framework calls for does not exist yet — and the same morning, before the announcement even went out, a letter signed by Geoffrey Hinton and two hundred researchers named that exact structure as the thing that must never happen: evaluations that are not meaningfully independent, with evaluators that carry significant other commercial business with the lab being evaluated. No one is accusing Accenture of being bought; the point is that the deal now obliges the world to care about the difference between access and independence. Access — employee-level, training-floor access — is delivered. Independence — funding, commercial separation, publication rights, protection from retaliation — is unaddressed. That gap, not the billion dollars, is the story: the industry has now commercialized the verification of the frontier before it has defined what verification is allowed to see, who pays for it, or who gets to publish what it finds. Watch the prospectus for Anthropic's answer, watch Accenture's earnings for the price of trust, and watch the letter's signatories for the standard that will eventually regulate both. Every dollar figure here remains a company announcement; the terms, when they come, will be the evidence.
Evidence labels used: CONFIRMED (joint official announcements, directly fetched; independent multi-outlet corroboration; the AI Evaluator Forum letter), COMPANY CLAIM (the "$1B each over five years" commitments and funding expectations), INDEPENDENT EVIDENCE (the AI Evaluator Forum letter's conditions, CASRAI's NIKOLAI analysis, market-move reports), INDEPENDENTLY VERIFIED (multi-outlet confirmation of the announcement itself), INTERPRETATION/PREDICTION (strategic readings, identified as such). Discovery-record inaccuracies (embedding direction, evaluation subject, "client-service product" framing) are corrected in Section 1.
