Trust Is Not a Press Release
The AI labs are spending real money to earn the public's trust. Whether it works depends on a question they can't answer for themselves.
The money is moving, and it is not small money.
In March, the OpenAI Foundation — the non-profit that controls OpenAI and holds a stake reported at roughly a quarter of the company, worth on the order of $130 billion — pledged to grant at least $1 billion within a year, across life sciences, jobs and economic impact, AI resilience, and community programs. On May 27 it named the first tranche: $250 million, aimed squarely at the economic disruption its own technology is expected to cause. Anthropic has its own Economic Futures Program — research grants, policy symposia in Washington and London, and an Economic Index built to measure how AI is actually changing work. Its CEO has said in plain language that the danger is not only lost jobs but "a level of wealth concentration that will break society."
I want to be clear at the outset: this is good. It is better than denial, better than silence, better than the usual posture of moving fast and letting someone else clean up. People of genuine goodwill are trying to get ahead of a problem before it lands. That deserves to be said, and said first.
But the money is the easy part. The hard part is the thing the money is being spent to buy, which is trust. And trust is the one thing an institution cannot purchase, announce, or award itself.
The structural problem with funding your own scrutiny
Consider what it means for a frontier lab to fund research into the economic damage its products may cause.
Even with the best people and the purest motives, the lab chooses the questions. It chooses the grantees. It chooses the measurement framework, the index, the definitions of "impact" and "transition" and "resilience." It convenes the symposium and writes the guest list. None of this requires bad faith. It only requires being human and being interested in your own survival.
The result, however well-intentioned, has a name, and the name is not trust. It is a better press release. A study commissioned by the disrupting party, framed by the disrupting party, and published under the disrupting party's banner is not an independent verdict on the disruption. It is the disrupting party's account of itself, with footnotes. The public has learned — through tobacco, through oil, through finance — to discount exactly this kind of self-authored reassurance, and it will discount this too, no matter how large the cheque.
This is not a cynical point. It is a structural one. You cannot certify your own claim. The referee cannot be on the payroll of one team. That is not a rule someone invented to be difficult; it is the condition under which a finding means anything at all.
What trust actually requires
If a foundation genuinely wants its money to earn trust rather than merely signal goodwill, there is a test, and it is uncomfortable by design.
The institution that receives the trust must be independent of the institution that wants it. Its process must be adversarial — not a panel of friendly experts nodding at a thesis, but a real contest in which someone credible is invited, on the record, to argue that the thesis is wrong. And the process must be able to produce a finding that goes against the party who would most like a favourable result. A verdict that can only come back "approved" is not a verdict.
Independence. Adversarial structure. A record that can rule against its own convener. Those three together are what distinguish a trustworthy institution from a well-funded one. They are also, not coincidentally, expensive in the only currency that matters here: the willingness to lose in public.
A worked example
I run a small one — deliberately small, deliberately independent. The Intelligence Commons holds public Hearings on contested claims about technology and the public interest. The format is adversarial and fixed: a Direct case, a Cross-Examination by someone recruited specifically to take it apart, a Re-direct, and then a finding by a panel named in advance — Certified, Failed, or Hung. The convener does not get to win by default. The record is the consequence, and the record can embarrass the person who built the venue. That is the point.
The current Hearing concerns the Work Contribution Continuity Framework — the proposition that as AI absorbs human labour, the fiscal and social obligations attached to that labour must follow the work rather than vanish with the worker. It is exactly the kind of claim a lab's foundation might like to see validated. So it is being tested the hard way. The Cross-Examination chair has been accepted by Robert D. Atkinson — founder of the Information Technology and Innovation Foundation and the author of The Case Against Taxing Robots — who has agreed to argue, on the public record, that the framework is wrong. The leading critic of the idea sits in the chair built to defeat it. If the framework cannot survive him, the record will say so, and that finding will stand.
No foundation paid for that. No lab framed it. That is the entire source of its value.
The standing offer
So here is the thing I would say to the OpenAI Foundation, to Anthropic, and to every lab and donor now writing cheques to earn the goodwill of the public and the commons they are reshaping:
The most trustworthy place for that money is not an institution you control. It is an institution you don't. Fund the independent venues — the ones that can cross-examine your favourite conclusions and publish a finding that costs you something. Underwrite the referee precisely because you cannot pick the score. That is what separates philanthropy that builds trust from philanthropy that buys coverage.
The civic infrastructure for this — adversarial, independent, on the record — does not yet exist at the scale the moment requires. It can be built. Some of it already is.
Trust is not a press release. It is a verdict you agreed in advance to live with, handed down by people you did not pay to agree with you. Anyone serious about earning it knows the difference.
Now that you know, what will you do?
Tom Tait is the founder of the Inter Species Wisdom Project Inc., a British Columbia Benefit Company, and convener of the Intelligence Commons. The live docket for the Work Contribution Continuity Framework hearing is here.