“AI-powered” is 2026’s most inflationary phrase, and it has quietly become a legal claim. Advertising regulators, competition authorities and securities frameworks all have tools against overstated AI capabilities, and enforcement has begun finding them. If your marketing, product pages or investor materials say AI does something it does not, you have three separate exposures wearing one buzzword.
Exposure one: advertising law
Türkiye’s Advertising Board (Reklam Kurulu) applies the standard misleading-advertising test with real fines and stop orders: objective claims require substantiation held before publication. “AI-powered”, “learns your business”, “99% accurate” are objective claims. If the “AI” is a rules engine, if the accuracy figure comes from a demo dataset, if humans in the loop do the actual work; the claim fails substantiation. The US FTC’s crackdown on AI washing signals where enforcement travels; the doctrine here needs no import, it already exists.
Exposure two: unfair competition
TTK’s unfair competition provisions give competitors standing against false capability claims, and in crowded AI verticals, a rival with a technical team that can reverse-engineer your product is a more motivated plaintiff than any regulator. Cease-and-desist letters citing benchmark claims are now a recurring genre in our inbox.
Exposure three: the fundraising deck
The same inflation in investor materials is not an advertising problem; it is a misrepresentation problem. Capability claims, pilot-customer counts and “proprietary model” language in a deck feed directly into the reps and warranties you will sign at closing (the ten diligence questions). The gap between deck and codebase is discovered in diligence at best; after closing at worst, when it becomes an indemnity claim.
The claims-substantiation file
For every public AI claim, hold one page:
- The claim, quoted exactly as marketing wants to run it.
- The mechanism: what the system actually does, in one engineer-approved paragraph.
- The evidence: benchmark, dataset, methodology, date; internal is fine, non-existent is not.
- The boundaries: what the claim does not cover, feeding the fine print.
- Sign-off: product + legal, dated, refreshed when the model or the metric changes.
Language that survives scrutiny
| Risky | Defensible |
|---|---|
| “AI eliminates errors” | “AI-assisted review reduced error rates by X% in our 2026 internal benchmark” |
| “Fully autonomous” | “Automates steps 1–4; human approval required for execution” |
| “Proprietary AI” | “Built on [model], fine-tuned on our proprietary dataset” |
The pattern: quantify, date, scope, and never let the adjective carry what the evidence cannot. Ironically, the substantiation file is also a sales asset; enterprise buyers increasingly ask for exactly this document under the name “AI fact sheet”.
This article is for general information only and does not constitute legal advice. It reflects the position as of July 2026.
Author
If this is on your desk
Templates and checklists are free in the Founder Academy; for a specific situation, book a 30-minute intro call.
Founder AcademyBook an intro call