Governors plan AI framework as Trump’s tech accord leaves enforcement to companies

Governors plan a framework, not a finished rule

Moore, who chairs the National Governors Association, said he would work with Braun, its Republican co-chair, and governors from both parties on the framework. He argued that states could not wait for federal action and said the White House meeting had not removed the need for the governors’ effort, The Associated Press reported.

The announcement was a commitment to develop a plan, not the adoption of one. The available account does not establish what standards the governors will propose, whether any would become binding or how states would enforce them. For people exposed to AI’s risks, those details will determine whether the effort creates a way to seek answers or remains another set of promises.

Governors are not starting from nothing. AP reported that Maryland, California, Illinois, Oregon and Virginia have taken steps on AI, while Utah and Texas have restricted data centers. But AP also noted the limits of state authority over technology that operates across state and national borders. A governors’ framework could bring state leaders together; its ability to set enforceable safeguards has not been established.

The accord’s four checks remain company-led

The accord sets out four layers of review. Companies are to put controls in place and monitor them, while an internal team checks whether those controls work. An independent external auditor or evaluator is to assess the safeguards, and a board committee is to review the results, according to CBS News’ account of the accord.

An outside assessment offers a check beyond a company’s internal team. Board review could also bring findings to executives with authority over company decisions. But the accord’s reported design leaves the process centered on the signatories: their controls, their internal reviewers, their outside assessments and their boards.

The reporting does not establish whether audit findings will be public. It describes no stated role in the review process for workers, customers or affected communities, and no route through the accord for them to seek accountability. Those omissions matter because a safeguard that identifies a problem is different from a requirement to disclose it or fix it.

Trump described the agreement as a moral commitment. “I think it's morally binding,” he said, as reported by CBS News. The accord leaves open the prospect of future legislation, but it does not itself create a law or state a penalty if a signatory fails to follow its steps.

The companies also agreed to meet regularly to develop safety standards and practices. That may help them identify common approaches, but the accord does not designate a public regulator to compel testing, require incident reports or penalize a failure to comply. Whether a company’s checks lead to changes remains an accountability question the document does not settle.

Federal enforcement remains a proposal

Trump said he planned to appoint an overseer and raised the possibility of an oversight committee. The available reports do not confirm that either has been put in place or describe enforcement powers for them. A proposed supervisor is not the same as an agency with authority to obtain information and impose consequences.

The case for the voluntary approach is that companies can build safety work into the development of more capable systems without treating innovation and protection as opposing goals. Nvidia CEO Jensen Huang argued that the two need not conflict and said companies are developing safety technology alongside their systems, CBS News reported. Anthropic CEO Dario Amodei acknowledged AI’s risks while saying how to address them remained under discussion.

Those positions explain why a company might support the accord’s internal checks and outside reviews. They do not answer what happens when a review finds a serious problem, or how people outside the company can learn whether its response was adequate. The distinction is between a promise to examine a system and a duty that someone independent can enforce.

Sen. Mark Warner, a Virginia Democrat, has called for Congress to require testing, evaluation and incident reporting for the most advanced AI models, rather than rely on voluntary standards, CBS News reported. His proposal points to a different source of accountability: requirements imposed by law. The available reporting does not establish that Congress has adopted them.

Trump also signed a separate executive order on federal terminology Tuesday, directing agencies to use Super Intelligence and SI in place of artificial intelligence and AI in certain non-statutory communications. That order does not make the companies’ safety pledges enforceable. The next concrete step described by state leaders is for Moore, Braun and other governors to develop their framework; what obligations it might propose remains unsettled.