Six out of 10 users of Zeta, a major South Korean AI character chat service, are teenagers or younger, data showed. On Crack, users aged 19 or younger accounted for close to half on a mobile app basis. Operators have introduced youth protection measures such as separating adult content and applying identity verification, but ensuring the safety of conversations generated in general content is emerging as a challenge.

MobileIndex data released on Sept. 8 showed Zeta had 1,586,178 monthly active users (MAU) in August. Users aged 19 or younger totalled 923,956, accounting for 58.3 percent. That number rose 29.3 percent in seven months from 714,390 in January.

Over the same period, Crack had 538,841 MAU, with 257,635 users aged 19 or younger, accounting for 47.8 percent. That was up 4.7 percent from 246,148 in January.

Zeta and Crack are leading services with large user bases in South Korea's AI character chat market. MobileIndex data showed average MAU in the first half of this year was 1.29 million for Zeta and 530,000 for Crack, ranking first and second, respectively.

Usage time was also long. Last month, average monthly time spent per user across all ages was about 49 hours 45 minutes for Zeta and about 26 hours 24 minutes for Crack. As of July, YouTube's average monthly time spent per user was about 44 hours 48 minutes.

Character chat is a service in which users hold long-running conversations with AI characters built around specific personalities and settings, focusing on role-play or relationship settings. Many characters and settings are premised on romantic or sexual relationships, and conversation intensity can rise even in general content depending on what users enter. Along with the service's long-term interaction characteristics, concerns include not only exposure to harmful content for youths but also excessive immersion and emotional dependence.

In a report published in July titled "Children and adolescents exposed to risks of AI chatbots", the National Assembly Legislative Research Service analysed that continuous interaction and always-on accessibility of AI chatbots could deepen emotional dependence or excessive immersion among children and adolescents. It said a preventive protection system is needed, including safe design and stronger operator responsibility.

Kim Na-jeong (김나정), an assistant legislative researcher at the National Assembly Legislative Research Service who wrote the report, suggested: "It is necessary to overhaul the protection system in a direction that strengthens safe design and operator responsibility, taking into account the structural risks of AI chatbots and the characteristics of children and adolescents."

Some voices also call for upgrading guardrails to prepare for situations where safety devices are neutralised through bypass inputs or "jailbreaks" in general content, separate from blocking access to adult content.

An AI safety industry official said, "Separating adult content and identity verification are basic protection measures, but character chat is a service where users and AI create conversations in real time, so it can develop in risky directions even in general chats." The official added, "It is important to safely manage the conversations that youths actually have with AI throughout the entire service process."

=== Overseas, limits on minors' chats; in South Korea, focus on separating adult content ===

Overseas, there are also moves to restrict AI chatting by youths itself. CharacterAI, a global AI character chat service, halted open chat features with AI characters for users under 18 on Nov. 11 last year. It had previously applied separate models, stronger filters and usage-time alerts for minors, but later tightened its policy to limit the scope of youth service use itself.

For age checks, it does not rely only on users' entered birth dates. It uses login information, platform activity and third-party signals. If a user believed to be under 18 seeks to confirm they are an adult, it also requires additional verification through an external provider.

South Korean companies say they operate youth protection measures centred on separating adult content, PASS identity verification and limiting exposure to general content.

A Scaterlab official, whose company operates Zeta, said, "As generative AI services and usage patterns change rapidly, user protection also requires ongoing checks and improvement." The official added, "We will continue to supplement safety devices based on cases identified in actual use."

A Wrtn Technologies official, whose company operates Crack, said, "In our own data, users in their 20s account for about half of the total, and teenage users are less than one-third the level of those in their 20s." The official added, "MobileIndex only counts app users, so it may differ from the actual age distribution."