Anthropic accuses Alibaba of mass AI theft via fake accounts

Anthropic has accused Alibaba of a large-scale attempt to steal its AI technology. The company claims attackers used thousands of fake accounts to interact with its AI model, Claude, millions of times. US regulators have since forced Anthropic to temporarily remove its AI software, Fable 5, from the market over security concerns. Anthropic states that nearly 25,000 fraudulent accounts linked to Alibaba made almost 29 million interactions with Claude. The goal was to gather intelligence on core aspects of its AI, including architecture and task execution logic. The attackers reportedly used a technique called model distillation to train new AI systems using Anthropic’s model outputs.

Regulators in the US have suspended Fable 5 due to fears it could be exploited for cyberattacks. Anthropic is now negotiating with these regulators over the suspension. The company is also locked in a dispute with them over alleged vulnerabilities in the software.

In response, Anthropic is urging Congress to take action. It wants US AI firms to have the freedom to warn each other about such attacks. The company is also pushing for stronger measures against Chinese companies involved in distillation. Additionally, it advocates for further restrictions on Chinese access to critical AI technologies, including advanced AI chips. The suspension of Fable 5 remains under discussion between Anthropic and US regulators. The company continues to press for legal changes to protect its technology. Its calls for tighter controls on Chinese access to AI resources reflect growing concerns over intellectual property theft in the sector.