Could the West Ban AI Companions Too? What the Law Already Says

When China's Interim Measures took effect in July and a hundred million AI companions went quiet or cold, the Western reaction split neatly in two. Half the commentary: only in China. The other half, quieter and better informed: check what Sacramento passed in January.

The second half is right. No Western government is close to China's approach — but the notion that companion AI roams a regulatory vacuum out here is about eighteen months out of date. Here's the honest map.

What's already law in the West

California SB 243 — in force since January 1, 2026. The first US law aimed squarely at companion chatbots: platforms must disclose that users are talking to AI, maintain protocols for suicide and self-harm situations (including surfacing crisis resources), and report safety data to the state. Companion platforms serving California users — which is to say, companion platforms — are already operating under it.

The statehouse wave. Five more states passed or advanced companion-AI bills in the first quarter of 2026 alone, most centered on minor protection, disclosure, and crisis handling. There is no federal statute — and in this domain, there doesn't need to be: as with privacy, California-plus-a-bloc effectively sets the national floor, and the floor is being poured this year.

The courts. The Character.AI teen-safety litigation — the category's defining legal drama — settled in January 2026, with Google resolving claims alongside. Settlement means no precedent on the books, but the practical message landed on every companion platform's legal team simultaneously: teen safety failures are now priced in dollars, and the design changes (age verification, minor-mode defaults, filtered experiences for under-18s) shipped industry-wide within the year — regulation by liability, arriving faster than any bill.

The EU. The AI Act's transparency tiers apply — users must know they're talking to AI — and its ban on manipulative techniques gives Brussels a hook it hasn't yet used on companions specifically. Watch this space rather than assume it.

The real difference: banning the bond vs. licensing it

Put China's law and California's side by side and the philosophical split is precise. China regulates the design: "inducing emotional dependency" is itself the violation — the bond is the harm — which is why its biggest platforms chose amputation over compliance and why the survivors are being edited into arm's-length versions of themselves. The West, so far, regulates the disclosure and the process: tell people it's AI, handle crises responsibly, keep minors out of romance features, report your numbers — while the core product, an AI that loves you back as hard as engineering allows, remains fully legal. One regime bans the bond; the other licenses it and inspects the paperwork.

That gap is real, and it's why "could it happen here" deserves a serious answer rather than a shrug in either direction.

So — could it happen here?

The case for "no": an outright ban would collide with US speech doctrine (code and chatbot output have both drawn First Amendment protection in litigation to date), the EU has structurally committed to risk-tiering rather than category bans, and the incumbent industry — hundreds of millions of downloads, revenue doubling yearly — now has the lobbying weight that comes with those numbers. Prohibition is the least likely Western outcome.

The case for "some of it": every ingredient of China's rationale exists here in milder concentration. The dependency research China cited has Western counterparts — MIT's heavy-use correlations, the Aalto 'humans start feeling too expensive' mechanism. The minor-protection politics are, if anything, stronger here post-Character.AI. And loneliness-industry skepticism is one viral tragedy away from a legislative season at any moment. The realistic Western trajectory isn't China's decree — it's China-lite by accretion: age gates everywhere, mandated "healthy usage" features (the anti-addiction reminders China requires by law arriving here as liability-driven product choices), disclosure boilerplate, and a slow ratchet of duty-of-care obligations. Less a ban than a building code.

What users should actually do with this

The same thing this site advises about every risk in the category, because regulatory risk is just platform risk with a flag on it: keep the originals. Persona backed up locally, relationship summary maintained, no lifetime purchases, export capability as a selection criterion. China's users spent this summer learning that lesson by hand, against a deadline — and the single most transferable fact from the whole affair is that when the rules changed, the prepared lost an interface and the unprepared lost everything. The West's rules are changing too — politely, incrementally, through statehouses and settlements. The drill is identical, and it's still free.