What did Korea's competition regulator publish on AI washing?

South Korea has drafted specific guidance against AI washing, the practice of advertising "AI" that a product does not really have. On 22 October 2026 the Korea Fair Trade Commission (KFTC), which enforces Korea's advertising law, published a draft Guideline on Labelling and Advertising Related to Artificial Intelligence and opened it for comment until 11 November 2026.

The draft explains when a seller may call a product "AI", what it must say about how that AI works, and when AI-made people and reviews in ads cross the line. A 60-page annex adds 16 before-and-after advertising examples, an eight-item self-check list and answers to common questions. The KFTC prepared it with the Korea Consumer Agency.

Why now? "AI" sells: in a Korea Consumer Agency survey from November 2025, 57.9% of respondents said they would pay more for an AI product, and those respondents would pay about 21% more on average. And AI-generated "experts", fake reviews and deepfakes are spreading in food and medicine ads on social media, harming older consumers in particular.

What is AI washing, and what counts as "AI" under the draft?

AI washing, in the KFTC's words, means advertising a product as using AI technology when it does not. The draft adds a second problem: using real AI but hiding its scope, operating conditions or limits.

The draft borrows its definitions from Article 2 of Korea's AI Basic Act (see the guide to the Korea AI Basic Act). The test is whether the product really uses learning, inference or judgment functions. Simple sensor-based control or threshold-based adjustment, which only gives a fixed response under preset rules, is not AI. On-device AI and AI delivered through a smartphone app can be.

The four unlawful types and the 16 examples

The draft sorts unlawful AI advertising into the four types in Article 3(1) of the Act on Fair Labeling and Advertising:

  • False or exaggerated (7 examples): claiming AI where there is none or where it plays only a minor role, or claiming expert-level ability without proof.
  • Deceptive (5 examples): hiding where real AI applies, its conditions or its limits, and unlabelled recommendations by AI-made people or reviews.
  • Unfair comparison (2 examples): "X times better" claims with no stated benchmark, or lab results presented as real-life results.
  • Disparagement (2 examples): calling rival technology "outdated" or "fake" while presenting one's own AI as "perfect".

The examples are invented but based on real files. Two show the approach.

The first is an electric heated mat advertised with "deep learning" that analyses the user's sleep in real time. In fact it has only a timer that, for example, lowers the temperature after three hours. The annex says the false AI claim, not the "perfect sleep" puffery, is the core of the problem.

The second is an "AI lawyer". The ad promises "the world's first AI lawyer" that lets anyone sue at the push of a button. In fact it runs on a general-purpose chatbot API, not specially trained on case law, with no review by lawyers and no quality comparison against real legal services. The annex treats the unproven claim of lawyer-level ability as false or exaggerated, and notes that using the title "lawyer" may also breach Korea's Attorney-at-Law Act. Legal tech as such is not the target: in 2025 the Seoul Administrative Court held that a template-based document service was not the unauthorised practice of law (Lawform v Seoul Bar Association), although it remarked that generative AI drafting documents from a user's account of the facts would fall within the ban.

The word "AI" itself is not banned: a robot vacuum with real object recognition keeps it, but must state next to the claim that the feature needs the app and may miss thin cables.

Virtual people, fake reviews and the eight-point checklist

AI-made people and reviews are treated as endorsements. Two annex cases stand out.

  • A virtual doctor selling food. An AI-generated "cosmetic surgery clinic director" claims an ordinary vitamin product reverses ageing. The annex points to a real case: in June 2026 the Ministry of Food and Drug Safety referred a food distributor to prosecutors over ads featuring AI-generated fake doctors. Amendments to the food advertising act (Law No. 21707, promulgated on 26 May 2026) and, according to the KFTC, to the Cosmetics Act and the Pharmaceutical Affairs Act expressly ban such virtual-expert ads from 27 November 2026.
  • A deepfake investment expert. A deepfake of a well-known investor promotes a paid stock-tips chat room with an unproven "300% average return".

The KFTC's endorsement guideline, as amended with effect from 1 June 2026, sets the labels: "[includes virtual person]" in text, "[virtual person]" on images and a permanent caption on video. The KFTC has already acted on manufactured reviews, for example in its 2024 Coupang decision on employee reviews, which is under appeal.

The eight-item checklist runs in four steps: is AI really used (and not a third-party model passed off as your own); are its scope, conditions and limits stated; are benefits described precisely; and can every claim be proved? Algorithms and parameters need not be disclosed.

Is the guideline binding, and what are the sanctions?

The guideline is not binding, but the law it explains is. It adds no new duties; it applies Korea's four advertising principles (truthfulness, completeness, fairness and substantiation) to AI, and failing a checklist item is not itself a breach. The Act has real teeth:

  • Proof: advertisers must be able to prove factual claims and file proof within 15 days of a request, or the ad can be stopped (Article 5).
  • Orders: stopping the ad, publishing the order, corrective ads (Article 7) and urgent temporary suspension (Article 8).
  • Administrative fine: up to 2% of the relevant sales, or up to KRW 500 million where there are no sales (Article 9).
  • Damages: liability does not depend on fault (Article 10).
  • Criminal penalty: up to two years' imprisonment or a fine of up to KRW 150 million (Article 17).

The draft is separate from the transparency duties in Article 31 of the AI Basic Act: those concern whether AI is used, the guideline whether it does what the ad says. AI labelling is handled separately (see Korea's AI labels move down the chain). Likewise, the Ministry of Science and ICT's AI product confirmation scheme, in operation since 21 July 2026 (Notice No. 2026-50, as cited by the KFTC), is evidence that AI is used, not a certificate of performance or a defence.

How do Türkiye and the EU regulate AI advertising?

Türkiye and the EU have no dedicated AI washing guideline like the KFTC draft, but both already ban the conduct through general advertising law. Türkiye also adopted specific rules on AI in ads this summer.

Türkiye

Article 61 of Consumer Protection Law No. 6502 requires truthful ads and, in Article 61(6), puts the burden of proof on the advertiser, like Korea's Article 5. The Regulation on Commercial Advertising and Unfair Commercial Practices bans misleading performance claims and scientific terms (Article 7), requires comparative claims to be proved with documents from universities or accredited or independent testing bodies (Articles 8 and 9) and bans disparagement (Article 10).

An amendment published in the Official Gazette of 1 July 2026 (No. 33297), in force since 1 August 2026, added two AI-specific rules and tightened the rules on reviews:

  • Article 18(8): where an ad uses AI or other software in a way that significantly affects consumers' economic behaviour, or shows AI-made digital characters that cannot be told apart from humans, this must be stated clearly.
  • Article 27(12): ads may not use an AI-made digital copy of a real person that falsely suggests the person used or recommends the product.
  • Article 28/B(8): buying untrue reviews or endorsements is banned, and under Article 28/B(6) reviews not verified as coming from actual buyers may not be used in ads.

The Ministry of Trade's Advertising Board (Reklam Kurulu) enforces these rules: it can stop or correct an ad, impose fines, suspend an ad for up to three months and order online content blocked (Article 77(12) of Law 6502). At its meeting of 12 September 2023 it ordered three ads to be stopped because their superiority claims, generated by ChatGPT, had no objective research behind them, according to the Ministry's announcement.

The gap is a definition. Turkish rules govern AI used in ads but have no published test, like the KFTC's, for when a product may be called "AI". An "AI lawyer" ad could also raise questions under Articles 35 and 63 of the Attorneys Act No. 1136, which reserve legal advice, representation and the title of lawyer to members of the bar.

European Union

The Unfair Commercial Practices Directive (UCPD) covers AI washing without naming it. Article 6 bans misleading information on "the existence or nature of the product" and its main characteristics, and Article 7 bans hiding material information, which would cover undisclosed operating conditions. Under Article 12, authorities can require proof of factual claims and treat unproven claims as inaccurate. Points 23b and 23c of Annex I blacklist unverified "real buyer" review claims and commissioned false reviews. In coordinated cross-border cases, maximum fines must reach at least 4% of turnover (Article 13(3)).

The EU has written specific rules for green claims in Directive (EU) 2024/825, which member states must apply from 27 September 2026; AI claims have no equivalent. For AI-made people, Article 50 of the EU AI Act requires deployers to disclose deepfakes (see the Article 50 kit and marking AI-generated content). Like Korea's Article 31, that is a transparency duty, not a test of whether the ad is true.

United States

In March 2024 the Securities and Exchange Commission settled charges against two investment advisers, Delphia and Global Predictions, for false AI claims, with penalties of $225,000 and $175,000. In September 2024 the Federal Trade Commission's Operation AI Comply targeted DoNotPay, which marketed "the world's first robot lawyer"; the company agreed to pay $193,000, and the FTC made the order final in February 2025. The KFTC annex cites this case, unnamed, beside its "AI lawyer" example.

What should businesses check before calling a product "AI"?

Businesses selling in Korea should test every "AI" claim against the draft now; the same checks suit Türkiye and the EU.

  • Document what the AI does and how performance figures were measured.
  • Where AI drives only a side feature, say which, as in "AI energy-saving washing machine", and put conditions and limits next to the claim.
  • Do not present a licensed model as your own development.
  • Label every AI-made person, voice or review, and never use an AI copy of a real person as an endorser.
  • In legal, medical and financial services, avoid suggesting that AI replaces a licensed professional.

What happens next?

Comments close on 11 November 2026. The KFTC will then consult other ministries, publish a final version with its start date, and update it as decisions accumulate. The Korean ban on AI-generated virtual experts in food, cosmetics and medicine ads starts on 27 November 2026. The first KFTC case under the final text will show whether the "learning, inference or judgment" test works in practice.

Frequently asked questions

What is AI washing?

AI washing is advertising a product or service as using artificial intelligence when it does not, or overstating what its AI does. Korea's Fair Trade Commission uses the term in its October 2026 draft guideline on AI advertising. It is generally unlawful under existing false-advertising law: Article 3(1) of Korea's Act on Fair Labeling and Advertising, Article 61 of Türkiye's Law No. 6502 and Articles 6 and 7 of the EU's Unfair Commercial Practices Directive.

What have the SEC and FTC done about AI washing?

The US Securities and Exchange Commission settled charges in March 2024 against Delphia and Global Predictions, two investment advisers, for false AI claims, with penalties of $225,000 and $175,000. In September 2024 the Federal Trade Commission's Operation AI Comply targeted DoNotPay's "robot lawyer" marketing; the company agreed to pay $193,000 under an order made final in February 2025.

Are there AI advertising laws in Türkiye?

Yes. Since 1 August 2026, Türkiye's Regulation on Commercial Advertising and Unfair Commercial Practices requires ads to disclose AI use that significantly affects consumers' choices and human-like AI characters (Article 18(8)). It bans AI copies of real people falsely endorsing products (Article 27(12)) and bought fake reviews (Article 28/B(8)). The Advertising Board enforces these rules.

Can a business call its product "AI" in Korea?

Yes, if AI really drives the feature advertised. Under the KFTC's draft guideline, the product must use learning, inference or judgment functions, not just preset rules or sensor thresholds. Where AI supports only a side function, the ad should say which one, and operating conditions and limits should appear next to the main claim.

Can ads use AI-generated people or reviews?

Only with clear labels, and never as fake testimonials. Korea's endorsement guideline requires "virtual person" labels, and from 27 November 2026 amended Korean laws ban food, cosmetics and medicine ads in which AI-generated virtual doctors or experts appear to recommend the product. Türkiye bans AI copies of real people as endorsers, and the EU blacklists false reviews and endorsements.

Burhan Doğuş Ayparlar's View

The KFTC draft creates no new duty, yet it answers the question every marketing team asks: when may we say "AI"? The "learning, inference or judgment" test is simple enough for a product manager to apply, and I expect other advertising authorities to borrow it.

Türkiye has the opposite profile: specific rules on AI used in ads, no test for AI claimed in ads. Given the Advertising Board's 2023 ChatGPT decision and the burden of proof in Article 61(6), no company should treat that gap as room to manoeuvre.

The "AI lawyer" example matters everywhere. A product can help people understand their legal position; an ad should not suggest that it replaces a lawyer's professional judgment.

Sources: the KFTC press release and draft guideline annex of 22 October 2026 (Korean; translations are the author's), Korea's Act on Fair Labeling and Advertising, Turkish Law No. 6502, the advertising regulation as amended on 1 July 2026, Attorneys Act No. 1136, the Ministry of Trade's announcement of 17 October 2023 on the Advertising Board's meeting of 12 September 2023, the Ministry of Food and Drug Safety notice of 7 August 2026 citing Law No. 21707, Directives 2005/29/EC and (EU) 2024/825, and the SEC and FTC releases cited (including the FTC's February 2025 release on the final DoNotPay order), all checked on 25 October 2026. The guideline is a draft and may change; its examples are invented. This is general information, not legal advice; the views are the author's.