Amodei says AI has not delivered yet. Buyers should write that into the contract.

Dario Amodei has agreed with his industry's critics. Writing on X on Saturday, Anthropic's chief executive said the most accurate criticism of AI companies, including Anthropic, is that "we haven't yet delivered on our big promises to benefit the world" ^[1], a concession that lands weeks before a listing his backers expect to value the company at $2trn in October ^[3].

The most useful line for anyone buying AI is not the admission but the schedule attached to it. Amodei said Anthropic hopes for incredible results in the coming years and some early glimmers in the coming months ^[4], which places the listing ahead of the glimmers ^[5]. He also drew the distinction that matters in a contract: saying AI will cure cancer has become a cliche most people read as deceptive, and the thing that will work is "actually curing cancer" ^[2]. That is the vendor's own test, applied to the vendor's own marketing, and it transfers directly to a statement of work. Roadmap language is a promise about the coming years. Acceptance criteria are about the coming months.

Treat the gap as a pricing input. If a $2trn valuation rests almost entirely on what the technology will do rather than what it has done ^[15], the enterprise buyer is being asked to fund the same expectation on worse terms, because a customer cannot exit a three-year commitment the way a shareholder can sell. The practical response is not to stop buying. It is to insist that anything described as forthcoming carries a date, a measurable output, a payment milestone tied to that output, and a termination right if it slips. Anything that cannot survive that treatment is a demo.

The public backdrop supports Amodei's premise rather than his optimism. Pew found half of American adults more concerned than excited about AI in daily life, against 10 per cent who were more excited, up from 37 per cent concerned when the question was first asked in 2021 ^[6]. That is a 13-point rise in concern ^[7] and a five-to-one margin against enthusiasm ^[8]. Amodei rejected the suggestion that his own warnings drove this, pointing to Machines of Loving Grace, his 2024 essay arguing AI could transform the world for the better ^[9], and attributed the distrust instead to decades of scepticism towards corporations, with AI as "just the latest iteration of it" ^[10].

That explanation was contested on the spot. An OpenAI employee replied that the pharmaceutical industry delivers advances constantly and remains among the least trusted sectors there is ^[11], and that AI companies will be judged on pricing, access, lobbying, opacity, how the gains are shared and who ends up holding the power ^[12]. Read as a procurement checklist, that is a better list than results alone: price escalation caps, capacity and rate-limit guarantees, model deprecation notice, audit and evaluation access. Yann LeCun, formerly Meta's chief AI scientist, argued the only viable path is AI that is widely available, shared and open ^[13], a position Anthropic rejects for its frontier models while Meta lobbies Washington the other way ^[14]. For a buyer, that dispute is a lock-in question: portability of prompts, fine-tunes and evaluation harnesses is the hedge, whichever side wins.

Watch whether the October listing prices at the expected level before any of the promised early glimmers arrive ^[3][4], and whether Anthropic attaches specifics to them. Watch renewal cycles for the first contracts written with milestone-linked payments rather than roadmap recitals.