AI Companion Rules Depend on Age Checks. Australia Shows Where They Break.
On July 15, China becomes the first country with dedicated national rules for AI companions, the apps built to remember you, stay in character, and keep you company. Five agencies, led by the Cyberspace Administration, finalized the rules in April. They ban virtual partners and virtual family members for anyone under 18, full stop. Children under 14 need a guardian’s sign-off even for the tamer companion services, and providers must build in time limits, spending blocks, and alerts to a parent when something looks wrong.
On June 15, the UK did something similar. Inside its plan to bar under-16s from social media by spring 2027, the government set a first-of-its-kind rule for AI. Romantic companion chatbots must enforce a minimum age of 18, and general chatbots must lock intimate features away from anyone younger. The public asked for this. In the government’s consultation, 94% of parents backed an age limit for chatbots, and so did 66% of children, more than three times the share of kids who wanted one for social media. Kids can tell these products are different.
Australia already ran the age-check experiment, and the results came back before China’s rules even take effect. Its under-16 social media restriction took effect on December 10. By March, platforms had removed or restricted about 5 million underage accounts. The regulator’s own compliance review found that as many as seven in 10 children who had accounts kept them anyway. A University of Newcastle study reported in June put it higher, with more than 85% of the under-16s researchers tracked still using the restricted platforms months in, and found the most common age check they met was a box asking for a birth date. Platforms let kids retry face scans until one passed, and some prompted users to bump their stated age past 16. In late June, Canberra answered by doubling the maximum fine to 99 million Australian dollars, about $68 million, and giving the regulator power to demand proof that the checks work.
Social media and companion apps are different products, and Australia is only a partial preview. A kid who slips through Instagram’s gate gets the same feed as everyone else. A kid who slips through a companion app’s gate gets a product whose whole job is the bond, with every protection meant for kids switched off. China’s rulebook names emotional dependence as the central risk, and nearly every safeguard it reserves for minors is keyed to knowing the user is one: guardian alerts, spending caps, youth modes, and crisis contacts. Beat the age check and the system files you as an adult. The rules also draft companions into crisis response, alerting a guardian or emergency contact when a user signals danger, a role I wrote about in April when two governments started treating chatbots as mandated reporters. That duty disappears too if the system thinks a 13-year-old is 30.
When Character.AI barred minors last October, the Bureau of Investigative Journalism found smaller developers set up shop to catch the displaced teenagers, advertising on Reddit and Discord, and some of those developers admitted their products were less safe.
There are honest counterpoints. Australia’s law penalizes platforms only when their safeguards fall short, and a court decides that. Verification may improve as regulators gain the power to audit it, and the UK says it will learn from Australia’s mistakes. China also brings enforcement muscle the others lack, because its internet already runs on real-name registration. And researchers at Oxford argue the sharper move is regulating the design itself, the features that make a bot feel human and keep people hooked, for every young user, whatever the app is called. That approach would not depend on a gate at all.
Australia can count the 5 million accounts its gates caught. It cannot see the children who beat the checks, because a child who slips past an age gate stops looking like a child to every system behind it. The new rules will protect the minors they can see. The ones most exposed are already filed as adults.