Summary：**OpenAI, Anthropic Sound Alarm: Chinese Labs Steal AI Using Fake Accounts***Introduction* OpenAI a

referrerpolicy="no-referrer"
style="max-width:100%;height:auto;display:block;margin:0 auto;">

**OpenAI, Anthropic Sound Alarm: Chinese Labs Steal AI Using Fake Accounts**

*Introduction*
OpenAI and Anthropic have issued a joint warning that Chinese research groups are exploiting tens of thousands of fabricated user accounts to harvest and replicate cutting‑edge AI models. The tactic, known as model distillation, allows attackers to train smaller, functional copies of large language systems without direct access to the original code or weights. As the practice spreads, industry leaders warn that it could trigger tighter export controls and reshape the economics of open‑source AI development.

*Key Developments*
According to internal telemetry shared by the two U.S. firms, suspicious login spikes began appearing in early 2024, originating from IP ranges linked to university labs and private enterprises in mainland China. The accounts, often created with automated scripts, queried public APIs millions of times per day, extracting output that was later used to fine‑tune surrogate models. OpenAI’s security team reported that over 120 000 distinct fake profiles were blocked in a single month, while Anthropic observed similar patterns affecting its Claude API. Both companies have since tightened rate‑limiting, introduced behavioral biometrics, and begun legal inquiries under the U.S. Export Administration Regulations.

*Industry Analysis*
Analysts say the incident highlights a growing asymmetry in the global AI race: while Western firms invest heavily in model safety and compute infrastructure, some overseas actors seek shortcuts through data scraping and account abuse. Dr. Lena Zhou, a policy researcher at the Brookings Institution, notes that “model distillation lowers the barrier to entry for capable actors, but it also erodes the incentive structure that funds frontier research.” The episode could accelerate calls for stricter API licensing, mandatory watermarking of generated