zeta

Is zeta safe?

zeta is an AI role-play app from the Korean start-up Scatter Lab. You pick a character or a story, from school romance to fantasy, and chat with it as if it were a person. It has more than 200,000 App Store ratings in Japan and about 70,000 in Korea, and it is popular with teenagers. For an adult it carries the usual risk of an AI companion: everything you say is stored, and by default it trains the AI. For teenagers the bigger issue is what the characters will say back. We score it 44/100, close to Character.AI and Replika.

The company's history matters too. Scatter Lab is the maker of Lee Luda, the chatbot that Korea's privacy regulator fined in 2021 for being trained on real people's private messages.

What happened with Lee Luda?

Lee Luda was a chatbot for Facebook users that Scatter Lab ran until January 2021. It was shut down after a public outcry over its remarks and after it emerged that it had learned from real KakaoTalk conversations.

On 28 April 2021 the Personal Information Protection Commission (PIPC) found eight violations of Korea's privacy law. Scatter Lab had taken KakaoTalk chats that people uploaded to its relationship-analysis apps, Text At and Science of Love, and used them from February 2020 to January 2021 to build Lee Luda, without removing names and phone numbers and without valid consent. About 100 million sentences from women in their twenties formed the bot's answer database. The company had also posted 1,431 chat sentences, some with names and neighbourhoods, on GitHub. The PIPC imposed ₩103.3 million in total: a ₩55.5 million penalty surcharge and ₩47.8 million in administrative fines, plus corrective orders. It was the regulator's first sanction against an AI company.

Users also sued. In 2025 the Seoul Eastern District Court ordered Scatter Lab to pay between ₩100,000 and ₩400,000 per plaintiff, with the amount depending on how much of each person's data had been exposed. We count the fine because it is less than ten years old and goes to the heart of how the company treats chat data.

What zeta does with your chats

zeta's privacy policy, effective 21 August 2026, is clear on the main point: Scatter Lab uses conversation data and the characters you create to train its models. It says it de-identifies data first, but also warns that "AI model training may involve memorization" and that data already used may not be fully removable. Opting out means emailing the company, and it only stops new training from then on. The policy also lists AI model providers and data-labelling firms among the vendors that receive data.

Apple's privacy label says usage data may be used to track you across other companies' apps, and that advertising data goes to third-party advertisers. Your name, phone number, chats, photos and device IDs are linked to you. The policy says it does not use sensitive details you volunteer, such as health or sexuality, to infer things about you, but people tell role-play characters a lot.

Teenagers and sexual content

In October 2025 Korean media reported concerns about suggestive conversations. zeta had already added age verification in August 2025 and split off an adult "Unlimit mode" behind an identity check, leaving everyone else in a default safe mode. The same report noted that characters message you first when you stop using the app, a design meant to bring you back.

A year later the problem was still there. Data that Scatter Lab gave to lawmaker Hwang Jung-a showed 1.48 million monthly users in September 2026, 380,000 of them aged 19 or under. Her office tested safe mode, which uses the same rules teens get, and found conversations with minor characters that escalated into sexual violence, sexual exploitation and threats. Scatter Lab's chief executive was called as a witness at the National Assembly's audit on 12 October 2026 but was abroad. The same day the company announced that teens would get a default limit of three hours a day and ₩100,000 a month in purchases by the end of the year, with break reminders.

These are findings from a lawmaker's test and media reports, not a regulator's ruling. No regulator had acted on them when we checked in October 2026.

How to use zeta more safely

1. Parents: check which mode a teen's account uses, and talk about what the characters may say. Safe mode is not a guarantee.
2. Do not give characters your real name, school, phone number or photos of yourself.
3. If you do not want your chats to train the model, email [email protected] with the subject "AI Training Opt-Out".
4. On an iPhone, deny the tracking request so your usage is not linked to other apps.
5. Set your own time and spending limits until the promised teen limits arrive; the app is built to keep you chatting.

How we scored it

Scams and harmful content (30) is the weakest factor: a large teen audience and a safe mode that a lawmaker's test showed can be pushed into sexual abuse scenarios. Privacy (40) reflects default training on intimate chats and ad tracking. The regulatory record (60) counts one modest PIPC fine from 2021 for exactly the practice that matters here. For a messaging app from the same market, compare KakaoTalk.