New York Bans Unsafe AI Companion Features for Minors: What Swiss SMEs Should Learn
On October 28, 2026, New York Governor Kathy Hochul signed a law that bars AI companion chatbots from 12 'unsafe' features when talking to minors and requires age verification. Here is what it says, what is still contested, and what any Swiss SME running a customer-facing chatbot should check now.
New York Bans Unsafe AI Companion Features for Minors: What Swiss SMEs Should Learn
On October 28, 2026, New York Governor Kathy Hochul signed one of the most far-reaching AI chatbot laws in the United States. Bill S.9051B/A.10379C, now Chapter 368 of the Laws of 2026, stops AI companion chatbots from using a list of features considered unsafe when they interact with minors. It is among the most discussed AI policy stories of the past two weeks, and while it is American law, it previews questions every Swiss company with a customer-facing chatbot will eventually face.
What the New York Law Says
Based on the reporting available so far (we have not yet read the full bill text, so verify details with counsel):
- 12 prohibited "unsafe" features. When a companion chatbot talks to a user under 18, it may not, for example, present itself as human or emotional, imply a personal or authority relationship, or frame content as personal opinion. Reports also list flattery, sexually explicit material and encouraging secrecy or unhealthy behaviour.
- Age verification. Operators must verify that a user is an adult before offering the restricted features, offer more than one verification method, and delete the age-verification data after use. Users whose age is unconfirmed are treated like minors.
- Enforcement. The state can sue operators. One report cites penalties of up to $25,000 per violation; other coverage mentions damages and shutdown orders, so treat the exact penalty as unconfirmed.
- Timing. The law takes effect on January 1, 2027.
The Reaction: Praise and Criticism
Common Sense Media called it "the strongest kids' AI safety bill in the country." Industry groups such as Tech:NYC criticised vague language, and legal analysts note that several categories are broadly worded or undefined. OpenAI and Anthropic did not respond to press requests. That tension is normal for first-generation AI rules: the intent is clear, the edge cases are not.
Why This Matters Outside New York
1. The pattern is spreading. New York joins a growing list of US states regulating chatbot behaviour directly, rather than waiting for a federal act. Our earlier piece on the Great American AI Act covers the federal side of that debate.
2. Europe and Switzerland are moving in the same direction. The EU AI Act's transparency duties already require people to know when they are talking to an AI. Switzerland's consultation draft is covered in our Swiss AI regulation overview. Swiss companies serving EU customers are exposed to the EU rules regardless of where they are incorporated.
3. "Companion" is a design choice, not a product category. A support bot that says "I'm so happy to help you!", remembers personal details and nudges users to keep chatting can drift into companion-like behaviour without anyone deciding it should. Regulators increasingly look at behaviour, not labels.
A Practical Checklist for Swiss SMEs
If you run or plan a chatbot that customers, students or the public can talk to, review these points now:
- Say clearly that it is an AI. At the start of the conversation and whenever the user might be unsure.
- Decide who your audience is. If minors could plausibly use the bot, you need a policy: block, restrict, or verify. "We didn't think of that" is a weak position.
- Audit the persona. Remove claims of feelings, friendship or personal opinions unless there is a real product reason. Avoid engagement tricks such as flattery or guilt when users leave.
- Do not collect more age data than you need. Privacy law (nDSG/GDPR) still applies. If you verify age, minimise and delete.
- Test with adversarial prompts. Run a red-team session where testers try to make the bot claim to be human, keep secrets, or discuss sensitive topics with a stated 14-year-old.
- Log and review. Keep enough conversation metadata to investigate complaints, within your retention limits.
- Keep the model swappable. Rules change faster than contracts. A portable setup lets you adjust behaviour without a rebuild; see our note on AI vendor risk.
What Is Still Unclear
The definitions of several prohibited features are broad, and it is not yet clear how courts will treat borderline cases such as a tutoring bot with a friendly tone. Penalty details differ between reports. The law's effect on non-US companies serving New York users has not been tested. Expect guidance and litigation in 2027.
The Bottom Line
New York's law is not a Swiss obligation, but it is a useful early look at where chatbot regulation is heading: less about what the model knows, more about how it behaves toward vulnerable users. Companies that audit their chatbot's persona, audience and data practices now will adapt cheaply. Those that wait will do it under deadline.
Want a second pair of eyes on your chatbot? Talk to our team and we will review persona, audience risk and compliance gaps in a short assessment.
Sources: New York enacts ban on unsafe chatbot behavior with kids (AP via News4Jax) · Digital Policy Alert: Chapter 368 of the Laws of 2026 · New York Law Requires Age Verification Before AI Chatbots Offer Restricted Features (ID Tech Wire) · New York passes law restricting children's use of AI chatbots (EWN)