Commentary. The most consequential question in artificial-intelligence policy is no longer whether governments will regulate AI. They are already doing so. The question is whether regulation will help people understand and challenge automated systems—or merely create a paper trail that companies can point to after something goes wrong.
Our view is that transparency rules are necessary, but insufficient. The European Union’s AI Act transparency obligations began applying on 2 August 2026, requiring, among other things, disclosure when people interact with machines and machine-readable marking of synthetic audio, images, video and text.[1] California’s AI Transparency Act also took effect this year, requiring generative-AI providers to offer watermarking, latent disclosures and detection tools.[2] These are reasonable first steps. They recognize a basic democratic principle: people should know when the apparent speaker, evidence or image in front of them was produced by software.
But a label is not accountability. A visible notice that content is synthetic cannot tell a job applicant why a system rejected them, whether a medical model was tested on people like them, or which company is responsible when an automated service causes harm. Detection tools can also be imperfect, particularly as generation and editing technologies improve. The risk is that transparency becomes a substitute for stronger safeguards rather than a foundation for them.
The case for restraint
There are serious arguments against a heavier regulatory regime. AI developers face fast-moving technical problems, and rigid rules can freeze yesterday’s assumptions into law. Smaller companies may struggle with compliance costs that large firms can absorb. Policymakers also risk imposing conflicting requirements across borders: the United States is developing a patchwork of state laws, while the EU is building a region-wide framework with its own timelines and categories.[2][3]
Those concerns deserve more than ceremonial acknowledgment. Excessive bureaucracy could reduce competition and push innovation toward the largest incumbents. A rule that demands perfect provenance for every generated word or image may also burden legitimate creative, educational and research uses. And no regulator can reliably predict every beneficial application of a general-purpose technology.
Yet the answer to uncertainty cannot be voluntary promises alone. A 2026 U.S. legal overview describes the country’s AI landscape as a patchwork in the absence of comprehensive federal legislation, while noting that the federal government has signaled a deregulatory and pre-emption-oriented approach.[2] That fragmentation may encourage experimentation, but it can also leave citizens with different protections depending on where they live and make it harder for responsible businesses to know what compliance means.
What meaningful transparency requires
First, disclosure should be paired with independent evaluation. Providers of high-impact systems—those used in employment, credit, health care, education, policing or essential services—should document performance, error rates, training-data limitations and foreseeable failure modes. Audits should be conducted by qualified bodies with access to enough information to test claims, not by vendors reviewing their own homework.
Second, individuals need a practical route to contest consequential decisions. A notice that an algorithm was involved is not a remedy. People should be able to request human review, correct materially wrong data and learn which category of information drove a decision, subject to legitimate privacy and security limits.
Third, regulators should enforce rules proportionately. Low-risk creative tools do not warrant the same obligations as systems that determine access to housing or public benefits. Clear thresholds, common technical standards and support for smaller firms would reduce compliance costs without abandoning public protections.
Transparency should answer not only “Was AI used?” but also “Who is accountable, what can go wrong, and what can I do about it?”
Industry groups are right that regulation must not become a brake on useful innovation. But the public interest is not served by treating speed as the only measure of success. Trust will not come from more labels alone. It will come when people can verify important claims, challenge consequential decisions and obtain redress when systems fail.
The emerging rules offer governments a chance to set that standard. They should resist both extremes: a loose promise that the market will police itself and a rigid rulebook that only the biggest companies can navigate. The durable approach is targeted oversight, independent testing and enforceable rights. Anything less risks making transparency a comforting word for an opaque system.