Anthropic says Alibaba launched largest-ever attack to copy Claude AI

The tech company Anthropic claimed that Alibaba was creating fake accounts to circumvent restrictions and get access to their Claude AI system. US-listed Alibaba stock fell about 2.7% to a 52-week low following reports of the allegations, extending the company’s broader 2026 decline.

This attack was described by the company as “the largest campaign ever to illicitly extract Claude’s capabilities,” raising fresh concerns over intellectual property protection and intensifying the AI rivalry between the United States and China.

The company further stated that, from April 22 to June 5, 2026, Alibaba used 25,000 fake accounts in over 28.8 million exchanges with Claude, thereby violating contractual terms.

In its June 10 letter to the US Senate Banking Committee, it also explained that Alibaba targeted Claude’s top strengths, specifically its advanced reasoning abilities, coding capabilities, and capacity to handle complex, multi-step tasks. 

Anthropic said the extracted data could potentially be used to accelerate development of competing AI systems. The accusations are likely to intensify the broader debate over AI security, export controls, and technological competition between Washington and Beijing, as governments around the world seek to balance innovation with the protection of advanced AI systems.

Anthropic had previously accused DeepSeek and MiniMax of distillation

Anthropic has characterized the campaign as a distillation effort that could accelerate China’s progress toward Mythos Preview-level performance.

In its letter to the committee, it called on lawmakers to close loopholes in China’s chip supply and hold labs accountable for distillation attacks. Nonetheless, it also acknowledged the steps taken by the US government to address these threats, particularly its partnerships with AI firms on intelligence-sharing programs. 

The Pentagon continues to include Alibaba on its list of Chinese companies with alleged affiliations to China’s People’s Liberation Army. Earlier this week, it had asked to be taken off the list, maintaining that it has no relationship with China’s PLA. 

For a while, US AI firms have warned that Chinese companies gaining their advanced tech threatens national security. Anthropic accusations are uniquely serious because they focus on Mythos, the exact model the US military and intelligence agencies have been deploying to launch cyber offensives against foreign adversaries. 

Previously, Anthropic also called out DeepSeek, Moonshot, and MiniMax for illegally using Claude’s responses to upgrade their own AI. It tracked 150,000 queries from DeepSeek, 3.4 million from Moonshot, and 13 million from MiniMax. At the time, it also asserted that distillation campaigns were becoming much more advanced, urging immediate teamwork across the global tech sector. 

Anthropic isn’t the only one pointing fingers—OpenAI has also called out Chinese groups for the same. It believes DeepSeek built up its platform by distilling its AI model.

Anthropic suspended access to Fable 5 and Mythos

Just last week, Washington ordered Anthropic to block foreign nationals from using its best AI models due to national security risks. Later, Anthropic confirmed that it had taken Fable 5 and Mythos 5 offline globally in response to a Trump administration directive banning foreign access.

According to the government, a technique exists that can override Fable 5’s protections and enable the model to identify software vulnerabilities. 

Nonetheless, a leaker on X under the domain synthwavedd noted that Anthropic may restore access to Fable 5. He said Tuesday that the latest version of Claude Code features text hints pointing to a Fable 5 revival, with weekly usage built right into subscriptions.

Anthropic originally intended to shift Fable 5 from a free preview to a separate, pay-per-use credit system. Instead, the possible update points to an integrated weekly quota that automatically refreshes. The leaker contended that removing the separate purchase disclaimer from the interface indicates a core change to the product’s billing logic rather than a simple string update.

In his post, he wrote, “The string ‘You’ve used your Fable 5 usage for this week’ has been added, and ‘purchased separately from your plan’ has been removed.”