Commentary: The United States is approaching a dangerous moment in the debate over artificial intelligence. The immediate risk is not only that lawmakers may regulate too slowly. It is that the rules governing one of the most consequential technologies of this era may be shaped by the companies with the greatest financial stake in avoiding them.
Recent reporting on the 2026 midterm elections found that political groups aligned with OpenAI and Anthropic had spent $55.7 million on advertising and voter outreach in key races. The companies do not share identical policy preferences: Anthropic has generally backed tougher safeguards, while OpenAI has favored a less restrictive approach aimed at speeding development. Yet both have an obvious common interest: electing policymakers sympathetic to the industry’s priorities.[1]
Our view is straightforward: companies building powerful AI systems should participate in the policy debate, but they should not be allowed to define its limits. Disclosure, independent testing and enforceable duties must come before political convenience.
The case for caution
Supporters of a light regulatory touch make a serious argument. AI development is global, and excessive American restrictions could push investment and talent toward countries with weaker safeguards. Competition with China is already central to Washington’s thinking, and the administration has resisted calls to slow development on those grounds.[2]
There are also practical concerns. Rules written before policymakers understand fast-changing systems can freeze outdated assumptions, burden smaller firms and protect established incumbents. A startup should not need the legal department of a multinational corporation to comply with every new requirement. Nor should regulators block useful applications in medicine, education or scientific research simply because they cannot predict every future misuse.
Those objections deserve respect. Regulation can be clumsy, and technological progress can produce genuine public benefits. The answer is not a blanket prohibition or a government attempt to select winners.
Why voluntary promises are insufficient
But the alternative now on offer appears too dependent on trust. The administration has created a voluntary framework for companies to submit models and tools for government assessment, while details of the framework remain secret, according to the BBC.[3] Voluntary participation may be useful as an interim measure, but it cannot substitute for public standards when companies face commercial pressure to release systems quickly.
AI failures are not confined to the laboratory. They can affect employment, credit, education, privacy, elections and public safety. A company’s internal review may identify some risks, but it cannot replace an independent process with transparent methods, appeal rights and consequences for negligence. The public should not have to accept that the same firms selling the technology are the only institutions capable of judging whether it is safe.
A better bargain
Congress should pursue narrowly tailored rules that focus on demonstrable harms rather than speculative claims about every possible future system. At minimum, developers of the most capable models should face mandatory incident reporting, independent evaluations, clear records of training and testing, and liability when they knowingly deploy systems that cause foreseeable damage. Smaller and lower-risk applications should face proportionate obligations.
Political spending also requires scrutiny. Campaign finance law should make it easy for voters to see which companies, executives and advocacy groups are funding candidates and advertisements. Influence is not automatically corruption, but hidden influence is incompatible with informed democratic choice.
Public confidence will not be restored by declaring AI either miraculous or apocalyptic. It will come from rules that acknowledge innovation’s benefits while refusing to outsource democratic judgment to private interests. The industry may help write those rules; it should never be permitted to write them alone.