Alibaba Bans Anthropic's Claude After Distillation Attack Claims
The Chinese tech giant cites security risks and will require employees to switch to its own AI assistant by July 10.
Alibaba will prohibit employees from using Anthropic's artificial intelligence tools starting July 10, marking a sharp escalation in tensions between the Chinese e-commerce giant and the San Francisco-based AI company.
The ban, confirmed by CNBC, comes after Anthropic sent a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs in June accusing Alibaba of conducting "the largest known distillation attack" on its systems. Anthropic claimed Alibaba "brazenly" and "illicitly" attempted to extract its AI capabilities, according to people familiar with the matter.
Alibaba has placed Anthropic's Claude Code on an internal high-risk software list. Employees must uninstall all Anthropic models and agent products and transition to Qoder, Alibaba's proprietary AI assistant. Both companies declined to comment on the ban.
Why it matters
This dispute illustrates the growing friction over AI technology access between U.S. and Chinese companies. As American AI firms tighten geographic restrictions, Chinese tech giants face difficult choices about compliance, workarounds, and developing domestic alternatives. The incident also highlights the vulnerability of AI models to distillation attacks, where competitors attempt to replicate capabilities by querying systems repeatedly.
Enforcement challenges and workarounds
Anthropic's terms of service explicitly ban Chinese companies and other "adversarial nations" from using its models. However, enforcement has proven difficult as Chinese firms have found ways to access Claude through third countries.
The Financial Times reported Friday that Anthropic is working to close these loopholes. According to the newspaper's sources, Ant Group provided employees with corporate Claude accounts accessed through the company's Singapore-based entity's intranet. ByteDance, TikTok's parent company, reportedly implemented a reimbursement program allowing engineers to expense personal Claude subscriptions accessed via virtual private networks.
A person familiar with ByteDance's policy told CNBC the reimbursement program, launched April 2, aims to help staff "experience and learn" about various AI products to enhance their skills. Ant and ByteDance both declined to comment on the Financial Times report.
Broader context
The ban follows widespread criticism in China directed at Anthropic after posts on Reddit and GitHub revealed hidden code designed to detect whether users might be based in the country. The technical measures reflect Anthropic's effort to enforce geographic restrictions, but they have sparked backlash among Chinese developers who view such detection methods as invasive.
The situation underscores the challenges AI companies face in policing model access across borders while maintaining competitive advantages in an increasingly fragmented global technology landscape.
These details were first reported by CNBC.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call