Tech companies cannot be their own AI regulators

Outside review could be valuable. But the accord does not specify how auditors will be selected, what criteria they will use or whether their findings will be public. Those omissions make it difficult for people outside the firms to judge what an assessment has actually tested.

Board oversight presents a similar problem. Directors may ask demanding questions, but a board committee cannot impose a public penalty for breaking this accord. If a company fails to carry out a promised review, the pact specifies no legal consequence. Its safeguards depend on the companies continuing to choose them.

Trump has argued against stifling AI development and praised self-policing. He also said he would name an overseer and proposed a committee of about 10 people to monitor the wider effort. AP reported Tuesday that those appointments were still to come. A proposed overseer, without defined enforcement powers, does not answer who could compel a company to comply.

AP reported that companies were already taking some measures in the accord or had promised them before. Continuing useful practices is better than dropping them. But restating voluntary commitments does not give the public a way to enforce them, however earnestly executives describe their intentions.

Innovation is not an exemption from accountability

The strongest defense of the accord is that safety work can improve without waiting for Congress. Nvidia CEO Jensen Huang said “there's no conflict between innovation, technology and safety,” CBS News reported. He is right that companies need not stop developing technology to test it. External assessments and internal teams should continue.

A poorly designed law could move too slowly as methods and risks change. That is a reason to write adaptable requirements, not to leave compliance optional. Public rules can set a floor while firms continue to improve their own checks above it. An auditor's review becomes more consequential when a public authority can demand information and act if a covered company fails to meet the rules.

Even an executive who signed the accord recognizes the need to confront risk. Anthropic CEO Dario Amodei said, “The technology has very real risks,” Euronews reported. He said the mechanism for addressing them remained under discussion. That is an argument for public deliberation over enforceable safeguards, not for letting the signatories settle the question among themselves.

The European Union's AI Act shows the difference between a promise and a rule with enforcement. For certain general-purpose AI model obligations, the European Commission's AI Office can request information or model access, require risk mitigation and, in specified cases, impose fines or restrictions. Those enforcement powers began applying Aug. 2, 2026. The act's scope and deadlines vary; it is not a single rule covering every AI system. But it establishes a public authority with powers the White House accord lacks.

Put the people facing the risks in the room

The case for public oversight is not confined to model testing. AP reported grassroots opposition across party lines to AI data-center construction. Trump said companies would provide local support or lower energy costs, but the accord sets no enforceable terms for that assistance. Communities weighing local costs deserve more than an assurance whose terms companies can decide for themselves.

The pact says its steps could later become law or regulation. That possibility is an admission of what the agreement cannot do now. It neither gives affected people a say in setting standards nor provides a public authority that can enforce the companies' commitments. Future legislation should address those gaps, not merely copy the accord's voluntary language.

Sen. Mark Warner, a Virginia Democrat, has called for mandatory testing, evaluation and incident reporting for the most advanced AI models, CBS News reported. Congress should enact those requirements, give a public authority power to obtain the information needed to enforce them, and provide a meaningful role for affected people in shaping the standards. Lawmakers can retain the accord's internal teams, external reviews and board attention. They should reject its central premise: that the companies subject to safety rules get to decide whether those rules have teeth.