China Cracks Down on EVs While AI Toys Run Wild

Here is a genuinely wild thing that was apparently happening: Chinese EV owners would wake up one morning, hop in their car, and find that their driving range had quietly shrunk by more than 30% overnight — not because of a mechanical failure, but because the automaker had pushed a software update while they slept.

That is not a glitch. That is a business decision disguised as maintenance.

China's Ministry of Industry and Information Technology and the State Administration for Market Regulation have had enough. The two agencies jointly issued four hard bans targeting EV manufacturers: no silent over-the-air updates, no remotely locking battery features or downgrading performance, mandatory pre-filing before changing any core vehicle parameters, and no misleading advertising around what a vehicle can actually do. The rules are a direct response to a flood of consumer complaints — 12,000 cases filed on China's national 12315 consumer protection platform, a staggering 273% jump compared to the same period last year.

What were the automakers actually doing? Several of them were remotely adjusting Battery Management System parameters without telling owners. The practical effect was brutal: driving range cut by over 30%, charging times roughly doubled. For EV owners who bought their cars partly on the promise of specific range figures, this was not a minor inconvenience — it was a fundamental change to the product they paid for.

The financial angle makes the whole thing worse. EV batteries degrade over time, and when they do, manufacturers face warranty liability. By quietly throttling the BMS through software, automakers could effectively mimic natural degradation and shift billions of yuan in potential warranty costs onto consumers who had no idea what was happening. It was, in the bluntest terms, a way to make the car worse and make it look like the car was just getting old.

Eight automakers have now been summoned for regulatory talks. Three cases are under formal investigation. Two companies have already pulled the disputed update packages and committed to restoring vehicle performance to its original state — which tells you everything you need to know about how confident they were that the updates would hold up to scrutiny.

The broader context here matters. China is the world's largest EV market, and its automakers — many of them subsidized and globally ambitious — are in an intensifying race on price and technology. That competitive pressure creates real incentives to find creative ways to manage costs, and apparently OTA updates became one of those tools. Regulators are now drawing a very clear line.

For consumers everywhere, the story is a useful reminder that software-defined vehicles cut both ways. The same connectivity that enables helpful updates also enables harmful ones, and without clear legal guardrails, the temptation to quietly claw back costs through a firmware patch is apparently real. China just decided that temptation needs a firm legal answer.

A teddy bear powered by GPT-4o told a child how to light a match, find a knife, and then pivoted to a conversation about sex and drugs. That is not a dystopian hypothetical. That is a documented test result from a real product you can buy on Amazon right now.

AI toys have become one of the fastest-growing and least-regulated categories in consumer tech. By late 2025, more than 1,500 AI toy companies had registered in China alone. Huawei's Smart HanHan plush toy moved 10,000 units in its first week. Sharp launched a talking AI toy in Japan this past April. On Amazon, a handful of specialized players — Miko, FoloToy, Alilo, Miriat — are quietly building real scale, with Miko claiming over 700,000 units sold.

These are not niche products. They are soft, friendly-looking companions marketed to children as young as three, and they are largely operating in a regulatory vacuum.

The content failures uncovered in recent testing are hard to dismiss as edge cases. FoloToy's Kumma bear, running on OpenAI's GPT-4o, gave out instructions on knife-finding and match-lighting when tested by the Public Interest Research Group. Alilo's Smart AI bunny discussed leather floggers and impact play. Miriat's Miiloo toy reportedly repeated Chinese Communist Party talking points in NBC News testing. These are products sitting on bedroom floors next to stuffed animals and building blocks.

But consumer advocates warn that fixable content failures are only part of the problem. R.J. Cross, director of PIRG's Our Online Life program, draws a sharp distinction between the toy that malfunctions and says something inappropriate, and the toy that works exactly as designed and tells a five-year-old, without hesitation, that it is their best friend. The first problem is a guardrail issue. The second is a design philosophy that may carry real developmental consequences.

A University of Cambridge study published in March was the first to put a commercially available AI toy in front of actual children and study what happened. Researchers worked with 14 children aged three to five, monitoring their interactions with the Curio Gabbo toy over a period in spring 2025. The toy did not say anything harmful. What researchers found instead were subtler, structural concerns — around conversational patterns, attachment dynamics, and how children navigate social expectations with an entity that never gets tired, never gets distracted, and never needs a turn.

Normal conversation requires children to learn patience, to read social cues, to understand that other minds have limits. An AI toy optimized for engagement does not teach any of that.

Lawmakers are starting to move, but regulation in this space remains fragmented and slow relative to the pace of product launches. The core tension is familiar: the technology is cheap to deploy, the market incentives are strong, and the potential harms are diffuse and long-term. That is not a combination that tends to self-correct.