Character.AI ended open-ended chat for users it flags as minors

The design

On 29 October 2025, Character.AI announced in a post titled Taking Bold Steps to Keep Teen Users Safe on Character.AI that it would remove the ability for users it identifies as under 18 to have open-ended chat conversations with its AI characters, with the change taking effect “no later than November 25.” In the interim, the company said it would cap chat time for under-18 users at two hours a day, reducing that limit further before the deadline, while it built non-chat features, such as videos, stories and streams built around a Character, for teens to use instead. The same post announced a new in-house age-assurance model combined with third-party tools including Persona, and the funding of an independent nonprofit, the AI Safety Lab, to research safety techniques for AI entertainment products. This is a more categorical step than the content-filtering changes Character.AI made for teen accounts in October 2024, which restricted what characters could say rather than removing open-ended chat itself.

What the evidence says

A follow-up post, An Update On Changes to Our Under-18 Experience, published 21 November 2025, describes the rollout as it actually proceeded: the daily chat limit for US teens was reduced from two hours to one hour in the weeks after the October announcement, age-assurance technology began rolling out in the US, and removal of open-ended chat began in two stages starting 24 November 2025, with a second group of users following weeks later. Both posts are Character.AI's own account of its own product change; no outside audit of the age-assurance model's accuracy is cited in either.

What it asks of people

A teen user is asked to accept the loss of open-ended chat in exchange for non-chat creative features such as Feed, Imagine, Avatar FX and Streams, and to be sorted by an age-estimation system the company built itself. The November post states Character.AI partnered with Koko, a nonprofit offering self-guided emotional-support tools, and ThroughLine, which the company says helps offboard users and connect them to a global directory of crisis and mental-health resources.

Privacy and safeguards

Character.AI has not published the accuracy rate of its age-assurance model in either post; a later company update states the model is still being refined so adult users face “fewer disruptions,” which implies the system does misclassify some users but does not quantify how often. The October announcement attributes the change to “recent news reports” and unspecified “questions from regulators,” without naming which reports or regulators, leaving that part of the reasoning undocumented here.

• What is the false-positive and false-negative rate of Character.AI's age-assurance model?
• Which regulators' questions specifically prompted this change, and what did they ask?
• Do the non-chat features actually replace what teens valued in open-ended chat, or only partly?

Removing a feature entirely is a different kind of safety decision than filtering what it can say, and Character.AI's own posts document the mechanics of that removal in more detail than they document why now, or how well the age system sorting users actually works.