Who Should Police AI Claims?
The DoNotPay case shows what can happen after an AI provider’s public claim outruns its evidence. This chapter asks a different question: which governance response is best suited to unsupported AI accuracy and capability claims? The supplied policy analysis compares no action, case-by-case enforcement under existing law, comprehensive AI legislation, and industry self-regulation. Comparing them by authority, timing, evidence, consumer stakes, and innovation makes the choice more useful than treating “regulation” and “innovation” as opposing camps.
Four ways to respond to a misleading claim
The first option is to take no action. A regulator could decide not to pursue an individual matter, and discretion can be reasonable when evidence is weak, the harm is remote, or another remedy is more appropriate. But treating non-intervention as the standing response would leave a predictable gap. Providers could make claims that invite reliance while consumers would have little reason to expect a public authority to challenge the mismatch. The report describes this risk in the context of the FTC’s consumer-protection role: systematic inaction could allow fraud or deceptive advertising to impose costs on consumers and honest competitors.
No action also sends a market signal. If a provider can claim professional-level accuracy without showing professional-level evidence, a competitor that spends money on testing and clear disclosures may be placed at a disadvantage. That does not prove that every disputed claim deserves enforcement. It shows why the cost of inaction includes more than one injured consumer: unsupported claims can affect how an entire market communicates capability and how consumers compare products.
The second option is case-by-case enforcement under existing law. This is the approach emphasized in the report and illustrated by the DoNotPay action. Under this model, the FTC examines a particular representation, audience, product, evidence record, disclosure, and likely consumer effect. The fact that AI is involved does not create an exemption from consumer-protection rules. A claim is assessed by the conduct and its consequences, not by the novelty of the technology alone.
Case-by-case enforcement has a practical advantage: it can adapt to different products and failure modes. The relevant evidence for a document assistant may differ from the evidence for a medical-information service, a financial tool, or a system marketed as a lawyer. An agency can focus on a concrete claim instead of trying to define every possible AI product in advance. It can also distinguish a provider that accurately describes a limitation from one that conceals a material objective or makes an unsupported comparison.
The same adaptability is a limitation. Enforcement usually follows a disputed practice rather than preventing every misleading claim before it reaches the market. Different cases may produce less uniform guidance, especially when products and marketing change faster than formal decisions. A small provider may also find it difficult to predict how an unusual claim will be evaluated until a comparable matter has been addressed. Case-by-case enforcement can therefore be proportionate without being complete. It supplies accountability for identified conduct, not a universal measurement standard for AI accuracy.
The third option is comprehensive AI legislation. A statute or broad regulatory regime could establish baseline duties, common definitions, disclosure obligations, testing expectations, or enforcement powers that apply across products. The attraction is legal certainty.