Anthropic Backlash Over Chinese Labs and Claude Claims

On February 23, 2026, Anthropic said Chinese labs including DeepSeek, Moonshot, and MiniMax used large numbers of fraudulent accounts to extract Claude outputs and improve their own models through distillation.

The claim quickly sparked backlash across developer and AI policy circles, with critics questioning evidence transparency and whether major labs are applying consistent standards to output reuse.

What Anthropic claimed

Anthropic said the activity involved:

• More than 16 million Claude interactions
• Roughly 24,000 fraudulent accounts
• Repeated attempts to capture reasoning and tool-use behavior

The company framed this as industrial-scale policy abuse and a safety risk if copied models remove guardrails.

Why the backlash escalated

Three arguments drove most of the criticism:

• Transparency gap: Anthropic published high-level numbers, but outside researchers cannot independently verify the full dataset behind the claim.

• Industry-wide distillation norms: Critics argue distillation is a standard optimization method, so the key dispute is where competitive learning ends and abusive extraction begins.

• Double-standard concern: Some developers say frontier labs defend broad data use for model training while drawing tighter lines once their own outputs are reused.

The dispute now goes beyond one accusation. It highlights a wider industry problem: policy language around distillation is still vague, while competitive pressure to reuse model outputs is increasing.

Related links

• Anthropic policy note: "Distillation by Chinese AI companies"
• TechCrunch coverage of Anthropic's allegations
• Reuters report (via Investing.com)
• Financial Times: Anthropic accuses Chinese labs of distillation attacks