Regulations
AI claim regulation sources
FTC enforcement records, SEC proceedings, and the ASA/CAP report each document what evidence was missing when regulators reviewed an AI claim. This page organizes those sources by claim surface so you can compare public AI marketing language against documented evidence gaps.
These sources record what regulators said was missing or unsubstantiated — not a current determination about any specific product, vendor, or claim. Not legal advice, a compliance certification, or a vendor approval list.
Claim sources
Each source has documented specific AI claim types where evidence was found missing. Case counts and claim types update as new cases are added.
AI marketing claims in product advertising — accuracy, automation, testimonials, and business outcome claims.
Most recent Cox Active Listening · May 2026
Investor-facing AI claims — trading platforms, investment advisor statements, and public company disclosures.
Most recent Presto Automation · January 2025
Advertising substantiation for AI claims — what 'AI-powered' must explain to meet advertising standards.
Most recent ASA/CAP AI as a marketing term · November 2024
Claim patterns and evidence questions
Each row maps a documented AI claim pattern to the evidence questions it carries. Evidence questions are drawn from the case materials listed.
| Source | Common claim pattern | Evidence questions | Related cases |
|---|---|---|---|
| FTC | Numeric accuracy, detection rate, or bias-free claim with a specific figure | Benchmark scope and design, item types tested, sample size, false positive and false negative rates, retest schedule | |
| FTC | Testimonials or review counts used as independent social proof | Reviewer independence, AI-generation or enhancement disclosure, collection timing relative to actual product use | |
| FTC | AI replaces a professional, eliminates human review, or automates a full workflow | Task scope tested, professional comparison standard, human-review boundary, escalation path when AI output is insufficient | |
| FTC | AI-linked earnings, income, or business outcome claim | Customer outcome distribution, total cost basis, time period, which result remains if the AI component is removed | |
| SEC | AI used for trading, investment decisions, or investor-facing capability statements | Model description and ownership, how AI contribution is measured and isolated, independent validation, disclosure document references | |
| ASA/CAP | 'AI-powered,' 'AI-driven,' or 'AI-enabled' used as a descriptor without explaining what AI does | What AI specifically does, what input it processes and output it produces, what the product would do without AI |
Cases by source
Selected cases from official or high-confidence sources, grouped by regulator. Each case documents a claim type, evidence gap, and buyer questions — not a verdict on any product or company.
FTC
- FTC Vague AI-powered
Cox Active Listening AI ad targeting
AI-powered targeting claim without evidence of data source, consent basis, algorithm function, or geographic accuracy
View evidence gap and buyer questions → - FTC Accuracy / Performance
Numeric accuracy without disclosed test scope or error rates
View evidence gap and buyer questions → - FTC Automation / Replacement
Professional-replacement positioning without task scope or qualified review boundary
View evidence gap and buyer questions → - FTC Compliance / Safety
FTC fake reviews and testimonials rule
Testimonials or review counts used as social proof without disclosure of AI generation, compensation, or collection method
View evidence gap and buyer questions → - FTC ROI / Outcome
AI-linked earnings claim without customer outcome distribution, cost baseline, or refund-promise substantiation
View evidence gap and buyer questions →
SEC
- SEC Automation / Replacement
SEC Presto AI voice automation
AI automation claim without disclosure of third-party technology, human intervention rate, or completion metric definition
View evidence gap and buyer questions → - SEC Accuracy / Performance
SEC Delphia AI investment claims
AI-driven performance claim without disclosed model description or independent validation
View evidence gap and buyer questions → - SEC First / Only / Best
SEC Global Predictions 'first regulated AI advisor'
Uniqueness claim without defined scope; AI capability claim without disclosed method
View evidence gap and buyer questions → - SEC Vague AI-powered
AI-driven investment platform claim without evidence of model ownership, live use, trading workflow, or investor disclosure support
View evidence gap and buyer questions →
ASA/CAP
- ASA/CAP Vague AI-powered
ASA/CAP AI as a marketing term
AI used as a marketing descriptor without explaining function, input, output, or user-visible benefit
View evidence gap and buyer questions →
How to use these sources
- Find the claim pattern in the matrix above. A specific number — '98% accurate,' 'detects all threats,' 'zero bias' — maps to FTC Accuracy / Performance. Claims about replacing a professional or eliminating human review map to Automation / Replacement. Claims in investor communications or financial product contexts map to the SEC rows. 'AI-powered' without any explanation of what AI does maps to ASA/CAP.
- Open a related case and read the evidence gap. The Workado case and NIST detector evaluation show what evidence to ask for around AI detector accuracy claims. The Delphia and Global Predictions cases show what the SEC found missing for investor-facing AI performance claims. Read the 'Evidence gap' and 'Buyer questions' fields in each case.
- Compare the claim wording you are reviewing. If the marketing copy uses a number without a benchmark, 'fully automated' without a human-review boundary, or 'AI-powered' without explaining what AI does, the evidence burden is likely similar to the documented case.
- Use /check to generate an evidence-burden note. Paste the URL or claim text into /check to get evidence requests, buyer questions, and wording boundaries for that specific claim.
Source updates and corrections
If a source status, date, company response, or case context is outdated, send a correction. Submissions are private by default and reviewed before any update appears on this page.
Send a correction →