Commentary. Artificial intelligence is moving faster than the institutions meant to govern it. That is not an argument for panic, nor for freezing technological progress. It is an argument for something more practical: enforceable rules, independent oversight and clear responsibility before increasingly powerful systems become embedded in decisions that affect safety, employment, privacy and democracy.
Congress is now debating whether AI companies should face direct federal regulation, whether existing agencies can manage the technology, and whether the United States needs a new specialist regulator. Proposals include independent testing, transparency requirements, privacy protections and mandatory reporting of serious incidents. The urgency is real: lawmakers and experts acknowledge that public oversight is struggling to keep pace with increasingly capable systems.
Our view is straightforward. The companies building these systems should not be allowed to serve simultaneously as developers, testers, risk assessors and final arbiters of harm. Voluntary commitments may be useful, but they are not a substitute for law.
The case for guardrails
AI systems can produce errors at scale, conceal the reasons for decisions and be adapted for harmful purposes. When a conventional product fails, responsibility is usually easier to locate. With AI, developers may blame users, users may blame the model and institutions may claim that no one could have predicted the outcome. That ambiguity is precisely why rules should establish duties in advance.
Independent audits would not guarantee safe systems, but they could test claims made by companies and expose serious weaknesses before deployment. Incident reporting would help regulators identify recurring failures rather than treating each one as an isolated mistake. Greater transparency could also give researchers, workers and affected citizens a meaningful basis for challenging automated decisions.
Regulation is particularly justified where AI is used in high-stakes settings such as health care, law enforcement, employment, credit or critical infrastructure. In those areas, a right to human review and an identifiable accountable institution should be minimum requirements.
The strongest counterargument
The opposing case deserves more than dismissal. Excessive regulation can protect established companies, raise barriers for smaller competitors and slow the development of tools that could improve medicine, education and productivity. Policymakers also worry that unilateral restrictions could weaken US technological leadership while rivals, especially China, move ahead.
Those concerns are legitimate. A rulebook written around speculative scenarios could bury useful innovation under paperwork. Regulators may lack the technical expertise to distinguish manageable experimentation from unacceptable risk. And forcing companies to disclose sensitive information could expose trade secrets without making the public safer.
But these objections argue for better regulation, not no regulation. Requirements should be proportionate to capability and risk. A small developer testing a limited tool should not face the same obligations as a company releasing a frontier model connected to millions of users. Regulators should protect confidential information while still demanding evidence that high-risk systems have been tested. Rules should be reviewed regularly, with public input and sunset mechanisms where appropriate.
Responsibility cannot remain vague
The central principle should be simple: power creates responsibility. Companies that build and profit from advanced AI are best placed to understand its design and foreseeable failure modes. They should carry obligations that users cannot realistically meet, including safety testing, documentation, security controls and prompt disclosure of serious incidents.
Government, meanwhile, must avoid writing rules so rigidly that they become obsolete as soon as the technology changes. A properly funded specialist office or agency could coordinate existing regulators, set technical standards and give the public a route to challenge dangerous practices. Congress should also preserve room for researchers and legitimate experimentation.
The choice is not between innovation and safety. Sustainable innovation requires public trust, and trust cannot be manufactured by press releases. It is earned when people know who is accountable, what protections exist and what happens when a system causes harm.
AI may become one of the most consequential technologies of this era. That is exactly why its governance should not be left to the speed of the market or the goodwill of individual companies. The sensible course is neither a blanket ban nor a regulatory vacuum, but independent scrutiny, targeted safeguards and consequences that make responsibility real.