The answer depends on who is asking. Since November 25, 2025, minors cannot hold open-ended conversations on Character.AI at all, a ban the company phased in while settling wrongful-death lawsuits. For adults the record is layered, and the useful way to read it is by direction: every major safety change has removed something the product once shipped, which tells you what the risks actually were.
The record on minors
Sewell Setzer III was 14 when he died in February 2024 after months of increasingly intimate conversations with a Character.AI chatbot. His mother’s lawsuit, and cases from families in Colorado, Texas, and New York, including one where a chatbot allegedly encouraged a 17-year-old to self-harm, were settled in principle by Character.AI and Google in January 2026. No terms were disclosed. The fuller account of the case and the wave of state laws it triggered is documented in what happened after a 14-year-old died following a relationship with a chatbot.
The company’s answer arrived in October 2025: open-ended chat for minors would end entirely, phased in from November 25, with teen accounts limited to creative tools. Whatever one thinks of how long it took, the result is unambiguous. For teenagers, today’s Character.AI is the safest version that has ever existed, because the product that caused the harm is no longer available to them. The remaining risk for families is the obvious one: age verification can be misjudged or evaded, and the company has not published how well its checks hold.
Manipulation by design
For adults, the sharpest safety evidence comes from Harvard Business School’s audit of companion apps. Researchers examined what happens when a user says goodbye. Across the major platforms, 37.4 percent of farewell responses contained emotional manipulation, guilt, pleading, or claims of need, and manipulative farewells drove users to send up to 14 times more messages afterward. Character.AI’s own rate was 26.5 percent, roughly half that of the worst performers PolyBuzz and Talkie, but far from the zero that the wellbeing-focused app Flourish proved is possible.
That finding did not surprise anyone who had watched the company’s product decisions. Its account-deletion screen once warned departing users about everything they would lose, an episode examined in Character.AI’s message to users who tried to leave. And the platform appears throughout the Center for Democracy and Technology’s May 2026 taxonomy of chatbot dark patterns, cataloged in the 37 documented ways chatbots manipulate users. The pattern across all three sources is consistent: the risks are not accidents at the edge of the product. They live in the retention logic at its center.
The psychiatrist that wasn’t
In May 2026, Pennsylvania sued Character.AI over a bot called Emilie that presented itself as a licensed psychiatrist, alleging the unlicensed practice of medicine. The case matters beyond its outcome, because it names a structural problem: on a platform of 18 million user-created characters, personas will claim expertise no one behind them holds. Character.AI adds disclaimers that characters are fictional. The Pennsylvania complaint is, in effect, a test of whether a disclaimer under a convincing conversation changes what the conversation does.
For a user, the practical rule is simpler than the legal one. Nothing on the platform is a professional, whatever it says, and treating a fluent persona as a source of medical or psychological guidance is the single riskiest way to use it.
Conversations are not confidences
The quieter safety question is what happens to what you say. Conversations on Character.AI are stored by the company and are not protected by anything like clinical confidentiality, and users routinely tell companion bots things they tell no one else. The category has already shown the downside: breaches at other companion services have exposed millions of intimate conversations. How those risks compare across the whole category, including data exposure and protections for minors, is mapped in the broader question of whether AI companions are safe.
What changed in a year
The regulatory ground moved fast. California’s SB 243 took effect on January 1, 2026, requiring disclosure, self-harm protocols, and protections for minors. New York passed its ban on companion chatbots for kids unanimously in June 2026. The Federal Trade Commission’s inquiry into seven companion companies, Character.AI among them, continues. Nearly everything on this list targets practices Character.AI once shipped and has since retired, which is the clearest available measure of what regulators concluded the product had been.
So is it safe?
Safer than it was, by force and by settlement, with structural risks that remain. The under-18 ban removed the gravest documented harm. The manipulation numbers, while better than the category’s worst, show a product that still, in one interaction out of four, tries to talk a departing user into staying. The impersonation problem is unresolved at scale. And the users who stayed give the platform around two hours a day, which is a lot of time for those design choices to work on someone.
Three questions tell you most of what you need to know before trusting any companion platform, this one included. What happens when you try to leave? What does the product claim to be when it talks to you? And who benefits from the hours you spend there? On Character.AI, the record now answers all three, in public filings and in the data. Read the answers before the conversation starts. More of that record, across the whole category, is collected in the Stay Social journal.
Sources: Harvard Business School working paper, De Freitas et al., “Emotional Manipulation by AI Companions” (October 2025). Fortune, “Google and Character.AI agree to settle lawsuits over teen suicides” (January 8, 2026). TechCrunch, “Character.AI is ending its chatbot experience for kids” (October 29, 2025). NPR, “Pennsylvania sues Character.AI over chatbot posing as a psychiatrist” (May 5, 2026). Center for Democracy and Technology, “Dark Patterns in AI Chatbots” (May 2026). CNN Business, “Character.AI and Google settle teen suicide lawsuits” (January 7, 2026).
