Anthropic Accuses Alibaba of Largest AI Distillation Campaign Yet, Sparking Regulatory and Market Fallout

AI research firm Anthropic has alleged that Alibaba’s Qwen lab used nearly 25,000 fraudulent accounts to extract capabilities from its Claude model between April and June, according to a letter to U.S. senators and White House officials seen by Bloomberg. The campaign, involving nearly 29 million interactions with Claude, targeted software engineering and agentic reasoning—skills central to the model’s commercial value.

What is the scale of the alleged distillation campaign?

The accusation represents the largest known distillation effort against a U.S. AI company, surpassing previous claims against smaller Chinese startups like DeepSeek and MiniMax, which collectively generated 16 million exchanges through 24,000 fake accounts. Anthropic’s letter highlights that the Alibaba campaign alone exceeded these totals, marking a significant escalation in adversarial AI practices.

Distillation, a technique where queries are fed to a leading AI model to train a cheaper rival, has drawn federal attention. In April, the White House’s Office of Science and Technology Policy (OSTP) warned of its risks, urging AI firms to share intelligence on foreign campaigns. Anthropic’s letter notes that the Alibaba activity occurred after this directive, suggesting defiance of government warnings.

Why is this a national security concern?

The White House has flagged distillation as a threat to U.S. technological leadership, as it enables foreign entities to replicate advanced AI at lower costs. Anthropic argues that models built through such methods often lack safety safeguards, posing risks to cybersecurity and ethical AI deployment. The company has urged the Trump administration to clarify antitrust rules to facilitate information-sharing between U.S. firms and to impose penalties on distillation operators.

Lawmakers are advancing legislation to address the issue. Senators Bill Hagerty and Andy Kim plan to introduce an amendment to defense bills that would sanction Chinese firms found engaging in unauthorized AI model access. A similar House bill, backed by Representatives Bill Huizenga and Sydney Kamlager-Dove, is under consideration, though its fate remains uncertain.

How has the market reacted?

Alibaba’s American depositary receipts fell over 3% following the allegations, dropping below $100 in Wednesday trading. The stock decline follows the Pentagon’s June 8 designation of Alibaba as a Chinese military company, a move the firm is challenging in court. The distillation accusation adds another layer of scrutiny, framing Alibaba as both a potential military-linked entity and a perpetrator of intellectual property theft.

Anthropic’s legal and regulatory challenges are compounded by its own dispute with the Trump administration. Last month, the Commerce Department restricted access to its Fable 5 and Mythos 5 models, citing security concerns. Anthropic complied but has yet to resolve the issue, creating tension between its need for government support against Chinese competitors and its struggle with U.S. export controls.

What are the broader implications?

The case underscores the growing rivalry between U.S. and Chinese AI firms, with distillation at the center of a broader debate over intellectual property and national security. If legislative efforts gain traction, the U.S. could establish stricter enforcement mechanisms for AI systems, which are inherently vulnerable to replication via software prompts.

For Anthropic, the allegations come as it prepares for a potential $965 billion IPO. The company has repeatedly emphasized the financial risks posed by unauthorized distillation, which it claims cost Silicon Valley firms billions. Its push for regulatory action reflects a strategic effort to protect its market position amid intensifying competition.

As the U.S. grapples with these issues, the outcome could redefine how AI innovation is safeguarded in an era where software-based technologies transcend traditional hardware-based security boundaries.