Commentary. The United States is drifting toward an AI regulatory system that is simultaneously too fragmented for the public and too uncertain for industry. That is not a reason to abandon oversight. It is a reason to make it coherent.
Across 2026, states have continued to legislate around chatbots, automated decision-making and frontier-model risks, while Congress has yet to pass a federal AI law. California has enacted more than 20 AI-related measures, including rules addressing transparency and synthetic performers, while Colorado replaced a landmark automated-systems law after concerns from business groups and litigation involving xAI and the Justice Department. The result is a growing map of different obligations for companies operating across state lines.
Europe offers a contrasting model. The EU AI Act’s general application date arrived on Aug. 2, 2026, giving European enforcement authorities powers over covered systems. At the same time, the EU’s AI Omnibus delayed some of the most demanding requirements for high-risk systems until 2027 and 2028. Even Europe’s more centralized approach therefore acknowledges a practical truth: regulation must be ambitious, but implementation cannot outrun the institutions expected to enforce it.
The case for stronger rules
Supporters of tougher safeguards have the stronger argument when AI affects a person’s livelihood, liberty or access to essential services. Automated systems already influence hiring, housing, finance and education. Errors in a recommendation engine are irritating; errors in a system that screens job applicants or evaluates eligibility for a public benefit can be life-changing.
Transparency is also a modest demand, not an anti-technology crusade. People should know when they are interacting with synthetic content or when an automated tool materially shapes a consequential decision. Independent testing, documentation and a route to appeal can help expose discrimination and make responsibility traceable.
The public has reason to be skeptical of voluntary promises. Companies face pressure to release products quickly, and safety claims made without meaningful audits are difficult for outsiders to assess. California’s new attention to synthetic performers and the EU’s machine-readable transparency rules reflect a sensible principle: users should not have to guess whether an apparently authentic image, voice or message was generated by a machine.
The case against regulatory sprawl
Opponents are right, however, that poorly designed rules can protect incumbents rather than citizens. Compliance costs fall most heavily on smaller developers, universities and nonprofit researchers that lack large legal and policy teams. Vague duties can also encourage companies to withdraw useful tools from certain markets instead of improving them.
There is a legitimate concern that lawmakers are regulating a moving target. A rule written for today’s chatbot may be irrelevant to tomorrow’s multimodal system. Colorado’s retreat from its earlier automated-decision framework illustrates how political and legal uncertainty can produce abrupt reversals. Businesses need predictable standards, while the public needs assurance that “innovation” will not become a blanket exemption.
Nor should every AI failure be treated as evidence that the technology itself is unacceptable. Existing consumer-protection, civil-rights, product-liability and privacy laws may already address some harms. Adding a new statute whenever a new application appears could create overlapping mandates without improving enforcement.
What a workable compromise looks like
Our view is that governments should regulate uses and risks, not marketing labels. A small entertainment model should not face the same obligations as a system used to determine medical treatment, employment or access to housing. High-impact uses deserve mandatory impact assessments, record-keeping, human review and a clear appeals process.
Rules should also be interoperable. State and national regulators should agree on baseline definitions, permit recognized third-party audits and publish technical standards that can be updated without rewriting entire statutes. Regulators need funding and expertise as much as lawmakers need restraint. A rule no agency can enforce is a press release, not protection.
Finally, policymakers should preserve room for research and contestability. Open scrutiny, independent evaluation and competition can reveal dangers that closed corporate testing misses. Security and privacy safeguards are necessary, but they should not be used to make powerful systems impossible to inspect.
The choice is not between an unregulated AI free-for-all and a bureaucratic freeze. It is between smart, proportionate guardrails and a patchwork that satisfies no one. The United States should choose the former before conflicting state rules become the country’s de facto national policy.