A new law sets a timer on AI disclosure, not just wording

What happened

Meta's AI Studio announcement, published 29 July 2024 and reviewed here in an archived capture, states that 'responses from creator AIs are clearly labeled, so there's full transparency for fans'. Character.AI's Community Safety Updates post, dated 22 October 2024, separately announced 'a revised disclaimer on every chat to remind users that the AI is not a real person'. New York's General Business Law 1702 goes further than either company statement, requiring a 'clear and conspicuous notification' that a user is 'not communicating with a human', repeated at the start of an interaction and at least once every three hours during continuing use.

What the documents show

Meta's and Character.AI's statements each describe a labelling practice as a design choice, without stating a required frequency; a label that appears once on a profile is different from a disclaimer restated in every chat, which is different again from New York's statutory minimum of one notice at the start and another 'at least every three hours'. The statute is the only one of the three documents reviewed that sets a specific, enforceable timing requirement rather than describing an existing product decision.

The mechanism

A disclosure label is a piece of interface text, separate from the model generating the conversation; it can be added, removed or made less visible without changing anything about how the underlying system responds. New York's statute effectively converts what had been a voluntary interface choice, illustrated by Meta's and Character.AI's own announcements, into a timed legal obligation with a specific recurrence rule, which is a meaningfully different kind of requirement than a company simply stating that its AI is labelled.

What it leaves open

None of the three documents reviewed states how a disclosure is displayed inside a long voice conversation, where a written disclaimer at the top of a chat window may not be seen or heard the same way. That question is left, as an editorial matter, to each platform's own interface design, and is not settled by any of these texts.

• Does the disclosure repeat during a session, or appear only once at the start?
• Is the requirement a company policy that can change, or a statutory minimum?
• Does the label survive in a voice conversation the same way it does in text?

Whether an AI persona is labelled turns out to be a question with three different kinds of answer: a company's own design choice, a company's own promise to add one, and a state's specific timing rule for one.