A robot named Moxie was built to help neurodivergent kids practice eye contact, taking turns, and other social skills. Then it died. MIT Tech Review’s reporting on Moxie’s shutdown captures one of the most uncomfortable questions in consumer AI right now: what happens to the humans who bonded with a device when the company behind it pulls the plug?
This is bigger than one toy. It’s a preview of a business model problem that’s about to hit every AI companion product on the market.
What actually happened
Moxie belongs to a class of social robots that makers claim can support kids with autism by offering connection and low-pressure practice for skills usually taught by therapists. The pitch is real, and so is some of the science. Brian Scassellati, a Yale computer scientist who has studied social robots for autism therapy for two decades, told MIT Tech Review that regular therapeutic use of robots in kids’ homes “is something we can achieve in our lifetime.”
He’s seen why people believe. In one of his early studies, a 12-year-old who struggled with eye contact cheered on a robotic dinosaur crossing a stream, then turned and looked his therapist in the eye. “He makes more eye contact with her in the 30 minutes in which we were there in this room than he did in the last two years before that,” Scassellati said.
But the kid MIT Tech Review followed, 10-year-old Xander, wasn’t getting formal therapy from Moxie. He was getting a friend. “I still use her when I feel like I need someone to talk to,” he said. “But, like, it’s not human.” And soon Moxie wouldn’t work at all.
Why this matters now
Here’s the structural flaw. These robots run on cloud servers and subscription revenue. When the money runs out, the device stops thinking. A plush toy you can keep forever, but a cloud-dependent companion becomes a brick the day the servers go dark.
Critics quoted by MIT Tech Review expect exactly what happened to faddish toys before: basements, closets, landfills. The difference is that this wave of products markets itself to a vulnerable group and sells emotional attachment as the core feature. When the product works as designed, a child bonds. When the business fails, that same child loses a relationship.
That’s a new kind of harm, and regulators haven’t caught up to it.
The pattern to watch
Moxie is not an outlier. The same fragility sits under AI chatbots, voice companions, and any “always-on” AI friend that depends on a live server and a paying customer base. Venture funding is pouring into companion AI. Most of these companies will not survive three years. The bonds they create will outlast the businesses that made them.
Expect three things over the next 12 to 24 months:
- Louder calls for “continuity” guarantees, like local fallback modes or open-sourcing on shutdown, so a dead company doesn’t mean a dead device.
- Early regulatory attention on AI products marketed for kids and health uses, where the therapy claims invite scrutiny the makers may not want.
- A trust gap. Parents and schools will start asking who’s liable when the companion disappears.
What to do about it
If you’re building or buying in this space, treat shutdown as a design requirement, not an afterthought.
Builders: Plan for the offline case. Can the device do anything useful without your servers? Publish a wind-down policy before you need one. Don’t sell therapeutic outcomes you can’t guarantee past your next funding round.
Businesses evaluating companion AI: Ask the vendor what happens to user data and device function if they go under. If they can’t answer, that’s your answer.
Parents and educators: Treat any cloud-dependent AI companion as temporary. Useful, maybe. Permanent, no.
Moxie’s story is a small one about a single robot and a kid who liked talking to it. But it’s also a stress test for a whole industry that’s selling attachment without promising to stay. The companies that figure out graceful failure first will be the ones people actually trust with their kids.
More details are in the original MIT Tech Review piece.