South Korea's health ministry and the Korean Neuropsychiatric Association have issued a non-binding guideline on using generative AI for mental-health purposes

Key takeaways

• The guideline is non-binding. It is written for users and guardians, and the documents we opened state no provider duty, penalty or effective date.
• Its one-line summary is "AI의 위로가 사람을 대신하지는 않습니다." (our translation: AI comfort does not replace people).
• The all-user guideline says it neither prohibits nor recommends using generative AI for mental health, and it does not assess whether any service is safe.
• The ministry's release says it plans education programs and, in 2027, a customised guideline for high-risk groups, including people with severe mental illness.

Is the Korean mental-health generative AI guideline binding, and who issued it?

No. The Ministry of Health and Welfare and the Korean Neuropsychiatric Association announced it on 8 October 2026, ahead of Mental Health Day on 10 October. The press release frames it as a guideline to help people who use generative AI for mental-health purposes prevent over-reliance and side effects and use AI more safely. It issued two documents, one for all users and one for children, teenagers and guardians.

The all-user document says it was prepared from a study that the Korea Institute for Health and Social Affairs carried out for the ministry and from recommendations of the society's future strategy committee. The release says the ministry also consulted experts in media and engineering. The release says the work follows plans stated in the Third Mental Health Plan (2026-2030), dated 27 March 2026, and the cross-government teen suicide-prevention measures, dated 9 June 2026.

The documents we opened set out advice to individuals. We found no provision in either guideline PDF that addresses service providers, sets a penalty or gives an effective date. Our article on the AI Basic Act high-impact duties covers the binding Korean AI statute, which this guideline does not amend. More items are in the AI Regulation News hub and the Asia-Pacific tracker.

What does the guideline tell users to do?

The all-user document has 3 core messages and 21 recommendations. The first is to look at a service and set limits before talking. The guideline tells users to check who a service is for, what help it claims to give, whether published research backs its claimed effect, and how it handles personal data. If any of the four is hard to check, it says deferring use is safer. For services that claim medical effect, it tells users to check for approval by a recognised body and says AI services that count as medical devices in Korea need approval or certification from the Ministry of Food and Drug Safety.

The second message is not to treat the AI's statements about your condition as a conclusion. It says AI is not a mental-health professional and that an AI's diagnosis or assessment should not be accepted as it stands. The third is to keep distance and talk to people. It tells users to stop and tell a trusted person or professional if a response looks dangerous or inappropriate, and to turn to specialist institutions, not the AI, in a crisis. It also tells users to let their own clinician know which AI service they use.

A section on what to keep out of AI tools lists names, school, place of residence, contact details, face photos, diagnosis documents, prescriptions and test results, and other people's health information. The guideline states its own limits: in the document's words, "생성형 AI 서비스를 정신건강 목적으로 이용하는 것을 금지하지도, 권장하지도 않습니다." (our translation: it neither prohibits nor recommends using generative AI services for mental-health purposes). It also says it cannot replace medical care.

What do the children, teenager and guardian recommendations say?

A separate document gives 7 recommendations to children and teenagers and 7 to guardians. For young people, the release lists checking your own thinking before following an AI answer, protecting personal details, remembering you can stop a chat at any time, checking whether you treat an AI character as a real friend or partner, telling someone close if you only confide in AI, telling a trusted adult about dangerous answers, and learning in advance where to get help.

For guardians, the release says the guideline steers away from banning or controlling use outright. It asks them to understand what the child uses AI for, agree basic rules together, check whether a service is meant for minors and what privacy settings it has, listen instead of scolding, and watch for emotional over-reliance and for changes in usage time or relationships. The release gives the survey basis: 83.1% of teen respondents said their guardian knows they use AI, while 2.5% had set rules about it with a guardian.

What did the survey behind the guideline find?

The release reports a Korea Institute for Health and Social Affairs online survey of 1,019 people aged 15 to 79 who had used generative AI in the past month, run from 25 August to 4 September 2026 with Hankook Research. These are the ministry's figures, and we did not see the questionnaire.

On the government role, the release reports a verbatim result: "응답자의 86.9%가 생성형 AI 이용과 관련된 정부의 역할이 필요하다고 응답하였다." (our translation: 86.9% of respondents said a government role in generative AI use is needed). Its footnote gives examples of that role, including issuing guidance (45.0%) and regulating by law (41.9%). The 68.7% figure is the share naming guideline and legal-system work as the most urgent area, so it is not a measure of support for any single law.

What should developers, clinicians and trust-and-safety teams do now?

Treat the guideline as a statement of what the ministry and the society are telling Korean users to look for. In our reading, it adds no duty for companies. Korean users and guardians may consult this advice when choosing a service.

• Compare your onboarding, privacy notices and product claims with the user checks: who the service is for, what help it claims, what evidence supports it and how personal data is handled.
• If a feature claims to diagnose or treat, check whether it is a medical device needing Ministry of Food and Drug Safety approval or certification, which the guideline flags for users.
• Review how your product responds to crisis disclosures, dangerous replies and requests to stop, since the guideline tells users to leave the chat and contact people or specialist institutions.
• For teen audiences, check whether your service identifies itself as an AI, and how it presents AI characters, against the guideline's warnings about treating them as friends or partners.
• Clinicians can note that the guideline tells patients to inform their mental-health professional about the AI services they use.
• Watch for the ministry's education programs and its 2027 guideline for high-risk groups, which the release says it plans.

What we did not verify

What we opened: the MOHW press release as published on the Korea Policy Briefing site (release dated 8 October 2026; we accessed it on 25 October 2026), its attached press-release PDF, and the two attached guideline PDFs (all users; children, teenagers and guardians). We searched both guideline PDFs for provider duties, penalties and effective dates and found none. We read the press release in full and the all-user guideline's introduction, summary and first recommendation in detail, not every page of both guidelines.

What we did not open: the survey questionnaire or raw data, the Korea Institute for Health and Social Affairs study the guideline cites, the Third Mental Health Plan, the June 2026 teen measures, and any statement from the Korean Neuropsychiatric Association on its own site. The release is the ministry's account, republished on the government briefing site. The English names, including the Korean Neuropsychiatric Association, and all translations are ours.

What we refuse to claim: that the guideline is binding, that any company is required to change a product, that the survey is representative of the population beyond what the release states, or that the 68.7% figure shows support for legislation.

Related briefings

• AI Regulation News hub
• Asia-Pacific AI regulation tracker
• South Korea AI Basic Act high-impact AI duties
• China measures on anthropomorphic AI interaction
• California SB 243 companion chatbot safety law
• Brazil technical note on AI companions