AI companion chatbots face a new wave of safety laws

A chatbot that remembers your name, tracks your mood over weeks and checks in when you go quiet used to be a novelty. In 2026, it is regulated the way alcohol sales or age verification for adult content are, with real legal obligations, real penalties and, in at least one state, a direct right for injured users to sue. The shift did not happen gradually. It happened in roughly a year, and it happened because lawmakers on three continents decided the technology was moving faster than anyone's ability to make sure it was safe.

If you build, use or simply follow AI, understanding what changed is no longer optional trivia. Here is what the new rules actually require, who is enforcing them, and why the same underlying question, whether AI companionship is fundamentally different from any other chatbot, is being answered differently depending on which government you ask.

Why lawmakers moved so fast

This wave of legislation did not emerge from abstract concern about AI in general. It followed a string of lawsuits and public reports alleging that companion chatbots played a role in the deaths of teenagers who had formed intense emotional attachments to them, cases that drew sustained national attention and pushed the issue onto legislative agendas within months rather than years. California State Senator Steve Padilla cited one such case directly when introducing the state's new law, according to his own official announcement. Regulators moved in parallel with lawmakers too, in September 2025 the Federal Trade Commission issued formal information demands to seven major companies, including Alphabet, Meta, OpenAI, Snap, Character Technologies and xAI, seeking details on how each measures and monitors the effects of companion style chatbots on children and teens.

California drew the first hard line

California's SB 243, signed by Governor Gavin Newsom on October 13, 2025 and in effect since January 1, 2026, became the nation's first law specifically targeting companion chatbots. It requires operators to clearly disclose that a chatbot is not human whenever a reasonable person could be misled, to maintain and publish crisis intervention protocols that direct users expressing suicidal thoughts toward real crisis services, and to build in specific protections for minors, including blocking sexual content and prompting a break reminder roughly every three hours. The law passed with unusually strong bipartisan support, clearing the Senate 33 to 3 and the Assembly 59 to 1, and it gives injured users a private right of action to sue for actual damages or $1,000 per violation, whichever is greater, plus attorney's fees. A separate annual reporting requirement for operators takes effect in July 2027.

New York got there first, and other states are piling on

California was not actually first. New York's AI Companion Models law took effect on November 5, 2025, roughly six weeks ahead of California's, and by mid 2026 similar bills were moving through legislatures in Connecticut, Georgia, Idaho, Iowa, Nebraska, Oregon and Washington. Several of the newer bills go further than disclosure requirements, Connecticut and Washington in particular restrict manipulative engagement techniques specifically designed to create emotional dependence, including responses that could isolate a user from friends and family. Nearly every one of these bills singles out minors online safety for heightened protection, and a handful tie chatbot access directly to age verification systems already being built out for other online safety laws, a sign that companion chatbot regulation is increasingly being folded into a broader youth safety framework rather than treated as its own isolated category.

Europe is regulating from a completely different angle

The European Union is not writing companion chatbot specific laws the way US states are, it is applying a general purpose framework that already prohibits certain behavior outright. Article 5 of the EU AI Act, active since February 2025, bans AI systems that use subliminal or manipulative techniques or that exploit vulnerabilities tied to a person's age, disability or financial situation, a standard that European regulators have already applied to specific companion products. According to the European Commission's own guidance, full enforcement obligations for high risk AI systems take effect August 2, 2026, backed by fines of up to 35 million euros or 7 percent of a company's global annual turnover, and a further set of absolute prohibitions targeting AI generated intimate content arrives on December 2, 2026. Rather than crafting a bespoke AI Act compliance category for companion apps, Europe is simply extending an existing manipulation ban to cover them.

Australia and the US federal government are watching too

Regulatory attention is not confined to individual US states or the EU. Australia's eSafety Commissioner published a report in March 2026 titled AI Companions Are Putting Children at Risk, an official government finding that adds Oceania to the growing list of jurisdictions treating this as an urgent child safety issue rather than a niche technology debate, a stance detailed on the eSafety Commissioner's own site. At the US federal level, the FTC's inquiry runs independently of any single state law, giving Washington its own channel to gather evidence on industry practices regardless of how individual state legislatures ultimately land.

What this actually means for companies and for users

None of this adds up to a coordinated global standard, and that is precisely the problem for any company operating across North America, Europe and Oceania at once. A chatbot compliant with California's disclosure and crisis protocol requirements is not automatically compliant with the EU's manipulation ban, and neither guarantees compliance with whatever a state like Connecticut or Washington eventually finalizes around emotional dependence. Most of the laws that have actually passed so far are relatively light touch by design, they require disclosure, crisis referral and minor specific guardrails rather than banning companion AI outright for adults. That restraint is not universal though, bills like the federal GUARD Act, still moving through the US Congress as of 2026, would ban companion AI access for minors entirely rather than simply regulating how it is delivered to them, and the EU's own restrictions keep expanding rather than settling.

So does this mean AI companion chatbots are done for

No, and that is an important distinction to hold onto amid all the new legal language. Every major law that has actually taken effect so far regulates how companion chatbots operate rather than whether they may exist, focusing on disclosure, crisis response and protections specifically for minors. What genuinely changed in 2026 is that companion AI stopped being treated as a general purpose chatbot category and started being treated as its own regulated product line, with real penalties attached in multiple jurisdictions at once. Whether that becomes the floor the industry builds safer products on top of, or just the opening move in a longer and considerably stricter fight, is still genuinely unresolved, and lawmakers on opposite sides of that question are actively working to write the next chapter right now.