The Federal Trade Commission published its proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems in the Federal Register on July 7, 2026, under docket FTC-2026-0859. The comment window closes July 31, a 24-day period. The statement applies Section 5 of the FTC Act to companies marketing AI systems as accurate, objective or neutral, and the reasoning reaches insurer chatbots, vendor-scored risk models and earnings-call language.
Key Takeaways
- "Consumers have a reasonable expectation that AI systems aim to give truthful and accurate outputs." That sentence carries the load; no misrepresentation of a specific fact is required.
- Hallucinations are not the target. Design decisions that steer outputs away from user-expected accuracy are, which makes this a gap between marketing and design rather than a model-quality rule.
- 24 days of comment window against the 30 to 60 days FTC statements of comparable scope have historically drawn.
- $193,000 in consumer redress in the 2024 DoNotPay settlement, the closest already-litigated precedent, out of the five actions in the Operation AI Comply sweep.
- 12-month fee caps are the standard vendor liability limit, with indemnities scoped to intellectual property and data privacy. Section 5 redress orders sit outside that structure.
What the Statement Actually Says
The proposed statement, filed under FTC File No. P264200, rests on the Commission's three-part deception framework: conduct is deceptive if there is a representation, omission or practice likely to mislead a reasonable consumer in a material way. Applied to AI, the position is that an operator steering a system's outputs away from the objectives users reasonably expect, without adequate disclosure, may be deceiving those users regardless of motive. It was issued in response to Executive Order 14365 of December 11, 2025, which directed the FTC to clarify how Section 5 applies to AI models.
| Item | Detail | Source |
|---|---|---|
| Federal Register publication | July 7, 2026 | Federal Register, doc 2026-13628 |
| Comment period close | July 31, 2026 | Federal Register |
| FTC docket | FTC-2026-0859-0013 | Regulations.gov |
| FTC file number | P264200 | Federal Register |
| Executive Order authority | EO 14365, Dec 11, 2025 | Consumer Finance Monitor |
| NAIC Model Bulletin adoptions | 24 states + DC (as of March 2025) | Quarles |
| Prior FTC AI action | Operation AI Comply, Sep 2024 | FTC press release |
The scope language is one sentence: "consumers have a reasonable expectation that AI systems aim to give truthful and accurate outputs." The Commission does not need to prove a specific misrepresentation about a specific fact. If an insurer, or a vendor operating on its behalf, has publicly framed an AI system as accurate, objective or neutral, and the design steers outputs away from that framing, Section 5 becomes available.
The distinction that governs how compliance teams should read it is not about output quality. Hallucinations arising from technical limitations are not by themselves within reach; design decisions that steer outputs toward objectives other than user-expected accuracy are, a point Covington's early analysis emphasized. Reading this as a rule about generative-AI hallucination misses it. It is a rule about the gap between what a system is marketed as doing and what it is designed to do.
The Documentation and the Marketing Are the Same Case
That gap has an actuarial location. The NAIC Model Bulletin on the Use of Artificial Intelligence Systems, adopted December 2023 and now in at least 24 states plus the District of Columbia, requires a written program covering model validation, testing and vendor oversight. That documentation is largely the same material actuaries produce under ASOP 56.
It is discoverable in an FTC investigation. If model documentation records a known accuracy limitation, a bias correction applied to satisfy a state fair-lending review, or a design decision to steer outputs toward a compliance objective rather than the user-expected one, and the carrier's marketing describes the model as accurate or objective, the Commission's file contains the internal actuarial acknowledgment of the design choice it is characterizing as deception. The gap between the documentation and the representation is the case.
Four surfaces produce the representations. Carrier-branded chatbots described on product pages as accurate, instant or objective sit closest to litigated precedent: the 2024 DoNotPay settlement produced $193,000 in consumer redress for marketing an AI chatbot as a "robot lawyer" without testing performance against the promised standard, inside the Operation AI Comply sweep that produced five actions.
Second, vendor-scored risk models. When a product page or investor deck describes a vendor module as producing objective or unbiased scores, the carrier has inherited a representation the vendor's design has to support. Third, AI-drafted claims explainer text: where a coverage-denial rationale carries a confidence signal the carrier's own governance flagged, omitting it is the potentially deceptive act. Fourth, earnings-call and investor-presentation language, which produces a durable public record in SEC-filed transcripts that an investigation can build on without subpoenaing marketing material.
The vendor layer is where the exposure is least priced. Contracts typically cap liability at 12 months of fees, with indemnities scoped to intellectual property and data privacy and reps limited to functionality and uptime. Section 5 redress orders and consent-decree conduct requirements sit outside all three, and the representation being challenged is the vendor's marketing rather than its product. Repeating a vendor's accuracy framing on a carrier's own page exposes both flanks at once, which is why the audit layer where carrier AI projects fail is now the layer where a Section 5 defense is built or lost.
The Defense Insurers Will Reach For, and Why It May Not Hold
The clean legal question is whether the McCarran-Ferguson Act, 15 USC sections 1011 through 1015, reverse-preempts Section 5 authority over the business of insurance. The statute reverse-preempts federal law that "invalidates, impairs, or supersedes" state insurance regulation, but only where no federal law specifically relates to the business of insurance. Section 5 is a general consumer-protection statute, which has historically produced a reasonably strong defense for carriers in areas covered by state regulation.
The statement is drafted to complicate exactly that defense. The Commission's preemption argument is that state law is impliedly preempted to the extent it conflicts with a federal regulatory scheme, and that a company's motives for deceiving consumers are irrelevant to the Section 5 analysis (Consumer Finance Monitor, July 2026). An insurer cannot cleanly assert that compliance with a state AI governance regime immunizes it, because the Commission has pre-committed to treating state-law compliance as irrelevant to deception. Whether courts accept that framing in an insurance context is open.
The dual-forum consequence is the one that changes an actuarial memorandum. State rate-filing governance for AI-driven pricing models already requires documenting model logic, validation and monitoring in the submission. A federal deception framework layered on top creates the possibility that documentation filed with a state department becomes the evidentiary foundation for a Section 5 case, which is a different risk profile from the one the filing was written against.
The balance-sheet consequence has no clean home either. An enforcement action runs through consumer-redress liabilities, class-action defense costs where private plaintiffs use consent decrees as templates, and rate-filing complications where regulators require disclosure of enforcement matters. None of those fits a standard extra-contractual reserve or a general expense accrual, and a carrier under active inquiry carries an emerging liability the loss triangles cannot see and ASOP 43 does not directly govern. The 24-day comment window, against the 30 to 60 days comparable statements have drawn, is the Commission's own statement about how quickly it intends the framework to be available.
Further Reading
- NAIC Third-Party AI Rules and What They Mean for Carrier Actuary Compliance: The state-side governance regime whose documentation becomes federally discoverable if the FTC opens a Section 5 inquiry.
- AI Pricing Model Drift and Rate-Filing Governance in P&C: The rate-filing environment now sits alongside the FTC framework as a dual-forum exposure for model documentation.
- Hartford's Algorithmic Impact Assessment and the Carrier AI Transparency Movement: The defensive posture carriers that have published transparency work carry into any Section 5 investigation.
- Why Carrier AI Projects Fail at the Audit Layer, Not the Tech Layer: The audit layer is where the Section 5 accuracy defense is built or lost.
- The AI Governance Gap in Actuarial Practice: The gap between actuarial model documentation and enterprise marketing is exactly the surface the FTC statement targets.
Sources
- Federal Register: Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems (July 2026)
- Regulations.gov: FTC Docket FTC-2026-0859-0013 (July 2026)
- GovInfo: Federal Register Volume 91 Issue 128 (Doc 2026-13628) (July 2026)
- Consumer Finance Monitor: FTC Takes Aim at AI Accuracy (July 2026)
- Inside Privacy (Covington): FTC Seeks Comment on Proposed Policy Statement Addressing AI Accuracy and Output Steering (July 2026)
- Spencer Fane: FTC Proposes New Policy on AI Accuracy (July 2026)
- FTC Press Release: FTC Announces Crackdown on Deceptive AI Claims and Schemes (Operation AI Comply) (September 2024)
- Quarles: Nearly Half of States Have Now Adopted NAIC Model Bulletin on Insurers' Use of AI (March 2025)