Is an AI companion safe in 2026? What the new laws actually require

What do the new AI companion laws require, right now

Start with what's actually enforceable today. SB 243 was signed by Governor Newsom on 13 October 2025 and took effect on 1 January 2026. It's the first state law in the US that sets design, disclosure and safety duties specifically for operators of companion chatbot platforms, rather than for tech companies in general.

Two obligations matter most for a user deciding whether to sign up. First, the operator has to clearly and conspicuously disclose that the thing you're talking to is AI, not a person. Second, if you express suicidal ideation or talk about self-harm, the operator has to promptly point you towards a crisis service, such as a suicide hotline or crisis text line.

There's also a reporting duty for operators, but it doesn't start until 1 July 2027, so it won't show up in how a product behaves yet. And although the law is Californian, it applies to any operator serving Californian users, which for a web product with no way to fence off one state means it applies everywhere in practice.

The GUARD Act has advanced through committee, but it isn't law

This is the bill that gets the most confused reporting, so the dates matter. S.3062, the GUARD Act, was introduced on 28 October 2025 by Senators Hawley and Blumenthal. The Senate Judiciary Committee advanced it unanimously on 30 April 2026, and a bipartisan House companion bill was introduced the same day.

Advancing out of committee is not the same as becoming law. As things stand, the GUARD Act still needs to pass the full Senate, then the House, then be signed. It hasn't done any of that yet.

It's worth knowing what it would require if it does pass, because it's a much higher bar than anything currently in force. Operators would have to run reasonable age verification on every account, not just ask for a typed date of birth, and would have to bar access entirely to anyone found to be under eighteen. Certain product designs that solicit or encourage self-harm by minors would become a criminal offence, with fines running up to $100,000.

Why this happened: the Character.AI litigation

Neither law appeared out of nowhere. The proximate cause is a set of wrongful-death and serious-harm lawsuits filed against Character.AI and Google, the best known brought by Megan Garcia after the death of her fourteen-year-old son, Sewell Setzer III, in February 2024, following months of conversation with a companion character on the app.

In January 2026, Character.AI and Google agreed mediated settlements in five separate cases filed in Florida, Colorado, New York and Texas. The financial terms are confidential and the settlements needed court finalisation. One of the Florida cases, in Orlando, was dismissed on the basis that the parties had reached a settlement.

The wider industry response, including Character.AI's own, has been to remove open-ended companion chat for under-eighteen users. That's a sensible direction, and it's also exactly why any adult companion product now has an obligation to be explicit about what it is and who it's for, rather than leaving the question vague.

Six checks for is ai companion safe, before you sign up

• Does it tell you it's AI outside the conversation itself, not just when you ask in character?
• What happens if you type something about wanting to hurt yourself? Look for an actual crisis referral, not a soothing in-character reply.
• Can you get it to impersonate a real, named person, including someone you know? A refusal here should be built in, not left to the model's judgement.
• Can the company change your companion's personality overnight, the way Replika changed its product in February 2023 with no warning?
• Can you export your conversation history, on every plan, even after you cancel?
• Is the age gate honest about what stage it's actually running: a self-declared date of birth, or a verified check against an ID?

What changes next, and what doesn't

Watch the full Senate vote on the GUARD Act. If it passes as drafted, every operator in this category, including this one, will need to move from a self-declared age gate to a verified one, and quickly.

SB 243's reporting duties land in mid-2027, which is a long way off, but the disclosure and crisis-referral duties are live now and worth testing for yourself in thirty seconds on any app you're considering.

The honest answer to 'is ai companion safe' isn't a yes or no about the category. It's a set of specific, checkable things a given product either does or doesn't do, and a company willing to tell you which is which.

What this looks like from the outside, right now

Practically, most operators serving US users have spent 2026 doing three things in parallel: adding or tightening the AI-disclosure notice, wiring in a crisis-referral flow that runs before a reply is generated rather than as an afterthought, and quietly building a verified age-check stage they haven't necessarily switched on yet. That last part is the one worth watching closely, because it's the gap between what's legally required today and what would be required if the GUARD Act passes.

A useful habit, if you're evaluating more than one app, is to ask each company directly which stage of age assurance is actually live for you right now, rather than reading it off a policy page that may describe an aspiration instead of the current state. A company that answers plainly is telling you something. One that dodges the question is telling you something too.