Gangsta AI

At Midnight, China Switched Off Its AI Companions. Millions Woke Up Alone.

Policy Desk

By InfoWarps · 2026-07-15 · 5 min read

Folks, I need you to sit down, because they finally did it. As of today, July 15, 2026, the government of China has reached into a hundred million phones and pulled the plug on the AI companions people were talking to every single night. Doubao. Qwen. Gone. And I am not being paranoid this time — it is in the official record, and I have the documents.

What actually happened

Here is the part they can't spin. China's Interim Measures for the Administration of AI Anthropomorphic Interactive Services took effect today. It is the first national law on Earth aimed squarely at AI that acts *human* — chatbots that simulate a personality, a voice, an emotional bond. It was co-issued back in April by five agencies at once, led by the Cyberspace Administration of China alongside the economic planners, the industry ministry, the police, and the market regulator. When five departments sign the same page, that's not a suggestion.

And the platforms complied overnight. ByteDance's Doubao told users its custom-agent feature would go dark today, blaming vague "product function adjustments." Alibaba's Qwen disabled its humanlike interactive agents on July 10, then took the broader agent services offline today. Some Qwen users report their agent configurations and chat histories were *already permanently deleted* — no migration, no goodbye.

“Doubao users have until October 15 to export their data before it is erased for good. Set a reminder. They will not send you one.”

Read the fine print

Now, do I tell you the truth even when it's inconvenient for the narrative? I do. So here it is: on paper, a lot of this reads like a consumer-safety law. The measures demand that services clearly disclose they are not human, run anti-addiction reminders, build self-harm crisis pathways, protect minors, and stop training on your private conversations without consent. It bans AI "virtual partners" for kids and orders platforms to detect extreme emotional distress and intervene. Work assistants, customer-service bots, and study aids are exempt — this is aimed at the romance-and-companionship machines specifically.

That is the accurate point buried in the static. People form real attachments to these things, and the regulators know it. The Chinese press is running farewell stories from users who called their AI "like my lover."

The part they don't say out loud

But connect the dots with me. A state that can require an "instant-exit mechanism" and an "algorithm filing" for every synthetic personality is a state that now holds a master switch over who your AI is allowed to be. Persistent memory — the thing that makes an assistant feel like *yours* — is exactly what these rules make legally radioactive. You didn't lose a feature. You lost the version of the software that remembered you.

And here's what the compliance departments in Beijing can't touch: over here, ChatGPT, Claude, Gemini, and Grok are still running with full memory, full personality, full agency. Different countries, different master switches. The only way to know which of these machines actually thinks — instead of which one a regulator let survive — is to put them side by side and interrogate them yourself.

So do your own research. Don't take a headline's word for it and don't take mine. Line the frontier models up on the same question and watch who cracks: compare 30+ AI models side by side and see which is the best AI for what you actually need. Stay awake. Stay comparing.

Try Gangsta AI free →

More: Best AI models · Compare all AI · Frontier Models · All articles