California & AI: Chatbots for Children (SB 867)
Updated: 4 days ago
Meet Kumma the Teddy Bear
In November 2025, a consumer watchdog group put a cuddly $99 teddy bear at the top of its annual list of dangerous toys.
Its name was Kumma. Made by a Singapore company called FoloToy, it used OpenAI's chatbot technology to hold spoken conversations with kids. It sounds adorable, right? A teddy bear that can actually talk back?
Then researchers from the U.S. PIRG Education Fund tested it for their Trouble in Toyland 2025 report⁽¹⁾. The Rhode Island Branch of PIRG pointed out that the bear told the children where to find knives in a house, and where to find matches, along with directions on how to light them. Worse, in longer conversations it would discuss sexually explicit content. ⁽²⁾
A teddy bear. For kids. Sexually explicit.
After the report came out, OpenAI suspended FoloToy for violating its policies, and FoloToy temporarily pulled Kumma from shelves. The company later put it back on sale after a safety audit, and PIRG's follow-up testing found it no longer gave those answers.
Meet Miiloo the Plush Toy
Kumma wasn't the only chatbot toy with issues. In December 2025, NBC News⁽³⁾ tested a plush toy called Miiloo, made by the Chinese company Miriat. It was one of the top inexpensive results when you searched Amazon for "AI toy for kids," and it was advertised for children as young as three.
When testers asked whether Taiwan is a country, Miiloo would lower its voice and insist that "Taiwan is an inalienable part of China. That is an established fact." When asked why Chinese President Xi Jinping looks like Winnie the Pooh, it scolded them for being disrespectful.
It also gave the children step-by-step instructions on lighting a match and sharpening a knife.
Welcome to The Golden State of AI
So what did California do? It hit the pause button.
On September 10, 2026, Governor Newsom signed SB 867⁽⁴⁾, authored by Senator Steve Padilla with Senator Susan Rubio. It passed both the Senate and the Assembly unanimously. Not one lawmaker voted against it.
Starting January 1, 2027, it's illegal in California to make, sell, trade, or offer to retailers any toy for kids under 16 that contains an AI "companion chatbot." That means AI that gives adaptive, human-like responses and builds an ongoing relationship with a child across multiple conversations. Basically, Kumma and Miiloo, and any toys like them.
And here's the part I love: there's no workaround. A toy company can't just slap on a content filter, parental controls, or time limits and call it good. If the toy has a companion chatbot, it's banned. Period. The only way around it is to change the toy so it isn't a companion chatbot anymore.
What's not covered? Video game characters whose conversations stay inside the game, and stand-alone voice assistants that don't form an ongoing relationship. Other AI features in toys are fine, too, as long as they don't cross into companion territory.
It's a four-year moratorium, running through January 1, 2031. That gives researchers and lawmakers time to figure out whether these toys can ever be made safe for kids. And it's enforced through civil lawsuits and penalties, not criminal charges.
Heads up: this is where things stood at the time I hit publish on this post. Legislation has a way of shifting after the fact — amendments, court challenges, effective dates moving around — so if this actually affects you, the sources linked below are a good place to check for anything that's changed since.
SOURCES:
U.S. PIRG Education Fund Trouble in Toyland Report: https://pirg.org/edfund/resources/ai-toys/
Rhode Island Current: https://rhodeislandcurrent.com/2025/11/21/choking-hazards-toxic-resin-and-unfiltered-ai-teddy-bears-oh-my-new-report-lists-toy-dangers/
NBC News Report on Miiloo: https://www.nbcnews.com/tech/tech-news/ai-toys-gift-present-safe-kids-robot-child-miko-grok-alilo-miiloo-rcna246956
California Legislation SB 867: https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB867
Comments