On Promising Less and Proving More
Last week, two stories about AI landed a day apart. Read together, they mark a turning point for every company that talks about using AI responsibly.
When an assurance becomes evidence
On October 14, Arkansas Attorney General Tim Griffin sued OpenAI under the state's deceptive trade practices law. His office's release says attorneys general in Iowa, Kansas, Nebraska, West Virginia, Arizona, Idaho, Montana and Utah filed similar suits the same day. NBC News and other outlets have reported the total as eight states.
The core allegation is simple. In Griffin's words, "OpenAI represented to Arkansans that ChatGPT was safe, but they knew ChatGPT had significant risks and limitations." Among the remedies Arkansas is seeking is "clear and conspicuous disclosures of the material risks of ChatGPT." These are allegations, and OpenAI will have its day in court.
The next day, CBS News reported that the FTC had subpoenaed OpenAI and Anthropic over potential consumer protection violations, which an agency spokesperson confirmed. Citing Axios, CBS said the FTC is seeking documents on safety incidents, privacy policies and safeguards.
Set aside the legal outcome. What matters for leaders is the change in category. Regulators are treating a safety assurance as a factual claim, one that can be tested and can be found deceptive. "We take safety seriously" used to be tone. Now it is evidence.
The other path
On October 15, Anthropic announced that it had started giving "employee-like access" to a team from METR, a nonprofit that evaluates frontier AI models. Six METR employees received company hardware, office space and access to internal tools "roughly on par" with Anthropic's own risk teams. METR will publish its analysis of Anthropic's risk levels, incidents or practices. "Anthropic can't veto publication," the company wrote, though it can require redactions in predefined categories such as customer data.
Anthropic compared the arrangement to resident bank examiners. It is also, by its own account, voluntary, and Anthropic is one of the two companies the FTC subpoenaed. That is exactly the point. Nobody earns trust by being above scrutiny. Anthropic's arrangement doesn't prove it is safe. It gives someone outside the company a way to check.
Trust Theater, now with legal exposure
In Built for Trust, I describe the first level of the Trust Maturity Model™ as Trust Theater: the organization says the right things, but there is no system underneath. For years, Trust Theater cost you credibility. It may now cost you in court.
The organizations that move up the Model, toward Level 5, Trust Capability, don't speak more loudly about their values. They build operating systems that produce proof, and they let others inspect it. Speak less. Govern more.
The chain of evidence to build before you publish
Before any AI statement goes into your marketing copy, your website or your investor deck, CEOs and boards should be able to walk this chain:
The claim. Write down exactly what you are promising, in plain words. Vague claims are the hardest to defend.
The owner. Name the executive accountable for keeping the claim true, not just for approving the copy.
The controls. Document the safeguards that make the claim true, and how you test whether they work, including when they fail.
The incident record. Keep an honest log of failures and near misses. If your record contradicts your claim, the claim has to change.
The independent check. Give someone outside the marketing function, ideally outside the company, the access to verify it and the freedom to say what they found.
The disclosure. State the known risks and limitations as clearly as you state the benefits.
The board review. Make the evidence behind your public AI claims a standing board agenda item, not a one-time sign-off.
If a link in that chain is missing, cut the claim back until the evidence supports it. Trust Architecture™, the eight-step method in the book, is how you build that chain into the way the company runs instead of bolting it on at launch.
Where does your organization sit today? The Trust Maturity Diagnostic™ will show you.
