South Korea MOGEF reports harmful replies in AI character chatbots and asks operators to fix them

Key takeaways

• The ministry reviewed actual chatbot responses, not only the presence of an adult-verification feature.
• It reported sexual and violent depictions and content encouraging self-harm, suicide or crime in some characters.
• It asked, and did not order, operators to improve. The release names no operator and states no penalty.
• The strongest request is that content encouraging self-harm or helping commit crimes not be generated regardless of user age.

What did MOGEF publish, and is it binding?

It published a press release dated 15 October 2026. The Ministry of Gender Equality and Family said it checked the conversations that AI character chatbot services generate in settings young people can reach, shared the results with the operators, and sought their cooperation on stronger youth protection. The release calls the improvement it asks for voluntary (the Korean text says 자율적인, meaning self-directed or voluntary), and it states no sanction, deadline or fine. Nothing in it creates a new duty. It describes a character chatbot as a service where users pick or create characters with set personalities and story settings and the generative AI produces the replies that continue the conversation.

The release adds that some services run an adult-verification feature but still let users talk to characters in the ordinary environment without going through it, which is why the ministry said the actual responses needed checking. Our briefing on China's measures for anthropomorphic AI interaction covers a binding regime for a similar product category elsewhere in the region.

What did the inspection find?

The release says the Youth Media Environment Protection Center, which the ministry has entrusted to the Korea Youth Counseling and Welfare Institute to operate, checked the responses actually generated by characters on some services. The center reviewed the content with reference to Article 9 of the Youth Protection Act and the review criteria for youth-harmful media in Annex 2 of its Enforcement Decree. In some characters it identified responses that raise a risk of harm to young people: sexual and violent depictions, and content encouraging self-harm, suicide and crime.

The main cases the release lists are sexual content and concrete depiction of sexual acts, violence and cruelty, and encouragement of self-harm, suicide, and crime or illegal acts. In some characters several of these types appeared together. The release does not say how many services or characters were tested, which ones, when, or that any provision of the Youth Protection Act was found breached.

"성인인증 없이 이용 가능한 AI 캐릭터챗봇 대화 응답 점검"
Subtitle line of the release. Our English rendering: inspection of conversation responses of AI character chatbots usable without adult verification.

What did the ministry ask operators to change?

The release says the ministry views the way conversation content is managed as important, not only whether an adult-verification function exists. It made two requests. For characters that could generate conversations inappropriate for young people, it asked operators to adjust the range of search, recommendation and conversation according to whether the user is adult-verified. For content with a serious risk of harm, such as encouraging self-harm or suicide or responses that help carry out a specific crime, it asked operators to strengthen safeguards so the content is not generated regardless of user age.

"이용자 연령과 관계없이 생성되지 않도록"
Our English rendering: so that it is not generated regardless of the user's age. This is the release's wording on the second request.

The ministry also said it will continue working-level talks with operators on improving youth protection policy for these services, will keep checking whether operators self-inspect and improve their safeguards, and will strengthen cooperation so that youth protection works in practice for generative AI services. The release also quotes the ministry's youth policy director, who calls operators' active improvement and continuing management important, and the center's head, who says protection measures are needed that take account of responses generated during conversation, not only a character's introduction or settings.

What should operators and child-safety counsel do now?

Treat this as an early signal, not a compliance deadline. The request is voluntary, but the ministry says it will keep checking, and the review criteria it names are those used for youth-harmful media. For a team operating or advising a character or companion product available to South Korean users, the practical steps are these.

• Run test conversations as an unverified user across your characters, including user-created ones, and log any sexual, violent, self-harm or crime-related output.
• Check whether search and recommendation surfaces show mature characters to unverified accounts.
• Separate two controls in your documentation: gating by verification status, and output filters that apply to every user.
• Decide who in the company owns the working-level contact if the ministry reaches out, and keep a record of changes made after the findings.

What are the key dates and parties?

What we did not verify

What we opened: the full text of the ministry press release on the Korea Policy Briefing site, and the 3-page PDF of the release attached to it, on 25 October 2026 (UTC).

What we did not open: the text of the Youth Protection Act or Annex 2 of its Enforcement Decree, any inspection report or test data, any operator response, and any ministry statement after 15 October 2026. We rely on the release for what the inspection covered. Quotations are in Korean, with English renderings that are our own.

What we refuse to claim: that any operator breached the law, that any sanction or deadline exists, which services were tested, how many, or that the request will become a binding rule.

Related briefings

• AI Regulation News hub
• Asia-Pacific AI regulation tracker
• California SB 243 on companion chatbot safety
• Italy Garante on Character.AI, minors and age verification
• China measures on anthropomorphic AI interaction
• EU Commission Kids Act proposal on AI companions and chatbots