Character.AI's terms disclaim reliance on what a character says

The design

Character.AI's Terms of Service, last updated by the company's own timestamp on 27 August 2025 and retrieved for this entry on 16 September 2026, is the contract a user accepts to use the product, not a description of one feature. Its conduct rules bar submitting content that 'seeks to provide medical, legal, financial or tax advice,' and a clause on third-party material states the company does 'not pre-screen content' and that AI-generated chats 'are unpredictable and may produce Generated Content that is inaccurate or offensive.' The same clause tells users they 'should not rely on the accuracy or completeness of statements made in content or Generated Content,' at their own risk. This is contractual language, not a court's or regulator's description of the product.

What the evidence says

A separate document, the company's own Community Safety Updates post of 22 October 2024, announced a different artifact: 'a revised disclaimer on every chat to remind users that the AI is not a real person,' among safety changes over the prior six months, pointing readers back to the Terms for its fuller policies. That in-app banner and the Terms' own reliance disclaimer are not the same document doing the same job. The banner is product copy shown inside a conversation; the Terms clause is a liability limitation written for a dispute, and nothing in the Terms text retrieved here uses the words fictional or not a real person. Treating the two as interchangeable overstates what the contract itself says.

What it asks of people

Accepting the Terms asks a user to give up more than attention to a chat banner. The document requires disputes to go through individual binding arbitration, waives a jury trial and class-action participation unless a user opts out in writing within 30 days, and caps liability at the greater of $100 or whatever the user has paid. None of that appears in the in-app banner the blog describes; it is visible only to a user who reads the Terms, called the entire agreement, superseding anything before it.

Privacy and safeguards

What the Terms protect is chiefly the company's own position in a dispute: a warranty disclaimer, a liability cap, and an arbitration requirement, alongside the narrower rule against seeking professional advice through the service. They do not verify that any in-app disclaimer displays consistently, define fiction for a sensitive topic, or address data retention. No regulator or court finding is cited here on how these clauses perform in practice.

• Does a safety announcement describe a user-facing feature, the company's contractual terms, or both, and does the wording distinguish them?
• What does a $100 liability cap actually protect a user against if a character's stated advice causes harm?
• Has a user considered the arbitration clause's 30-day opt-out window, if that option matters to them?

Read together, the two documents describe two different promises: one a product-surface reminder aimed at a person mid-conversation, the other a liability shield aimed at whoever might later sue. Neither substitutes for the other, and an announcement about one should not be mistaken for a change to the other.