Alibaba has banned its employees from using Anthropic’s AI tools, effective July 10, citing security risks. This decision follows Anthropic’s accusations that Alibaba conducted a ‘distillation attack’ to illicitly extract its AI capabilities, violating terms of service that prohibit use by ‘adversarial nations’ like China.
The move also comes amidst controversy over Anthropic’s hidden code designed to detect users in China and efforts to close loopholes allowing Chinese companies, like Ant and ByteDance, to access Claude via third countries or personal subscriptions. Alibaba will now require employees to use its own AI assistant, Qoder.
READ MORE FROM CNBC
Chinese e-commerce giant Alibaba is taking decisive action, prohibiting its employees from utilizing Anthropic's artificial intelligence tools for work-related tasks, effective July 10. The company has cited concerns regarding potential back-door security risks associated with the U.S. AI developer's products.
Sources familiar with Alibaba's internal operations, who requested anonymity, confirmed that Anthropic's 'Claude Code' has been placed on a high-risk software list. This move comes after Anthropic's recent accusation against Alibaba in a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs.
Anthropic alleged that Alibaba 'brazenly' and 'illicitly' attempted to extract its AI capabilities, labeling it as the 'largest known distillation attack' on its systems to date. This incident is particularly sensitive given Anthropic's terms of service, which explicitly prohibit Chinese companies and other 'adversarial nations' from accessing its AI models.
Alibaba employees are now mandated to remove all Anthropic models and agent products from their systems. Instead, they are directed to use Qoder, Alibaba's proprietary AI assistant. Both Alibaba and Anthropic have declined to comment on these developments.
The ban coincides with growing online criticism in China directed at Anthropic. Discussions on platforms like Reddit and GitHub have highlighted the presence of hidden code designed to detect if users are based in China, sparking further controversy.
The Financial Times reported recently that Anthropic is actively working to close loopholes that have enabled Chinese entities to circumvent restrictions and access Claude via third countries. The newspaper, citing anonymous sources, indicated that Chinese fintech group Ant had provided its employees with corporate Claude accounts, accessed through the company's intranet connected to its Singapore-based entity.
Furthermore, ByteDance, the parent company of TikTok, reportedly does not directly facilitate access to Claude but has initiated a reimbursement program. This program allows engineers to expense personal subscriptions to various AI products, which they can then access using virtual private networks (VPNs).
Ant and ByteDance refrained from commenting on the Financial Times' report. A source familiar with ByteDance's internal policies, who preferred not to be named, stated that the reimbursement program, launched on April 2, aims to encourage staff to explore and learn from a diverse range of AI products to enhance their skills.