Commentary. The United States is approaching an avoidable mistake on artificial intelligence: treating the choice between innovation and regulation as if it were a binary one. The current congressional stalemate may be politically understandable, but it is not a responsible governing strategy. As increasingly capable systems move into workplaces, schools, hospitals and public services, the absence of clear national rules leaves citizens to negotiate risks one lawsuit, agency memo or corporate promise at a time.
The case for action is not hypothetical. A recent assessment of the American debate describes new AI safety legislation as unlikely in the near term, even as calls for Congress to establish a federal framework grow louder. The result is a patchwork: states are developing their own oversight, while companies face uncertainty about which standards will apply and people affected by automated decisions often lack a clear route to challenge them. The BBC reported on the congressional deadlock.
Our view is straightforward: Washington should establish a limited national floor, not attempt to micromanage every model or freeze development. That floor should require meaningful testing for high-risk systems, disclosure when people are interacting with machines, independent scrutiny of systems used in consequential decisions, and a usable process for correcting serious errors. It should also preserve human responsibility. “The algorithm did it” cannot become an excuse for an employer, insurer, hospital or public agency.
The strongest objection deserves respect
Critics are right about one danger: badly designed rules can entrench the largest technology companies. Compliance costs may be manageable for firms with vast legal and engineering departments but prohibitive for startups, universities and smaller developers. Vague safety mandates could also invite regulators to suppress legitimate research, slow useful products and give foreign competitors an advantage.
There is a second, more practical objection. AI changes faster than legislation. A statute written around today’s chatbots could be obsolete before agencies finish implementing it. Some policymakers therefore favor voluntary standards, targeted enforcement under existing consumer-protection laws, or a temporary federal pre-emption of state rules to create one predictable market.
These arguments are not excuses for inaction; they are design requirements. Rules should be proportionate to risk rather than based on the mere presence of AI. A spellchecker and an automated system deciding whether someone receives medical care should not face the same obligations. Congress should pair requirements with technical assistance for smaller organizations, protect good-faith research, and review thresholds regularly. Federal consistency can be useful, but it should not erase stronger protections where states have acted to address concrete harms.
Why voluntary promises are insufficient
Industry participation matters, and companies possess expertise government cannot easily replicate. Yet voluntary commitments cannot substitute for enforceable duties when commercial incentives reward speed, scale and market share. Even technology executives have recently urged stronger safeguards, including outside evaluation before release and mechanisms to pause systems when human control is compromised. TechCrunch reported those calls for global safety rules.
The answer is not to accept every proposal from industry leaders. Companies may favor rules that their largest rivals can afford, or standards that focus on dramatic future risks while neglecting familiar harms such as discrimination, fraud, privacy breaches and opaque workplace surveillance. Independent civil society groups, workers, creators and affected communities must have a voice in writing and enforcing the rules.
International coordination is equally necessary. AI systems cross borders, and incompatible national standards will create both loopholes and friction. But a global agreement cannot become a waiting room in which no government acts. The United States can set a domestic baseline while cooperating on shared testing, incident reporting and protections for human decision-making. Brookings has likewise argued for federal rules with international scope and human-centered safeguards.
Congress should therefore reject both complacency and panic. The objective is not to stop progress, nor to guarantee that every model is harmless—a promise no regulator can make. It is to ensure that those who build and deploy powerful systems remain answerable for foreseeable consequences. Delay may preserve flexibility for Washington, but it does not preserve safety for everyone else.