China’s AI agent rules force Doubao and Qianwen to axe smart personas

China’s new rules on “AI anthropomorphic interaction services” are an early, concrete test of how far governments will go in regulating AI agents that feel like people. Forcing Doubao and Qianwen to shut down user‑built agents isn’t about model weights or benchmarks; it’s about emotional attachment, parasocial relationships and high‑frequency, low‑value traffic that regulators see as both socially risky and economically dubious. That’s a very different risk frame from the Western focus on cyber‑offense and disinformation, and it signals that China may be willing to intervene directly in product design, not just data and compute.

Strategically, this nudges Chinese AI platforms away from C‑end, UGC‑heavy “smart companions” and toward B2B and enterprise agents where revenue per token is higher and regulatory exposure is lower. For the global race, it means Chinese labs may double down on industrial, logistics and robotic agents rather than open consumer ecosystems of personalized bots. That could slow Chinese progress on the most human‑centric interfaces, but accelerate tightly integrated, domain‑specific agency.

If other jurisdictions copy this logic, the path to AGI‑like assistants embedded in everyday life could bifurcate: constrained, heavily‑regulated companion agents in some markets, and more free‑wheeling agent ecosystems in others, which in turn shapes where the toughest alignment and misuse problems actually emerge.