Anthropic CEO Dario Amodei confirmed that the company has officially cut off access to its AI model, Claude, for all firms linked to the Chinese Communist Party (CCP).
Amodei revealed that Anthropic chose to forgo “several hundred million dollars” in revenue to enforce this ban, which targets entities designated by the US Department of War as Chinese Military Companies and any organization with more than 50% Chinese ownership.
The “Distillation Attack” Evidence
The decision follows a massive investigation by Anthropic that uncovered what it calls “industrial-scale” intellectual property theft by three major Chinese AI labs: DeepSeek, Moonshot AI, and MiniMax.
- The Scale: These labs allegedly created over 24,000 fraudulent accounts and used commercial proxy “hydra clusters” to bypass regional blocks.
- The Volume: They generated more than 16 million exchanges with Claude to “distill” its reasoning, coding, and agentic capabilities into their own models.
- The Target: Anthropic noted that when a new model was released, these firms pivoted nearly half their traffic to the new system within 24 hours to extract its latest capabilities.
The National Security Rationale
Anthropic has framed this blockade as a necessary defense of “Democratic AI”:
| Risk Category | Anthropic’s Concern |
| Safety Guardrails | Anthropic argues that while Claude has strict filters against helping develop bioweapons or offensive cyber tools, these safeguards are stripped away during the distillation process by CCP-linked firms. |
| Export Controls | By reverse-engineering Claude, Chinese firms are effectively circumventing U.S. export controls on advanced chips intended to preserve a technological lead. |
| State Compulsion | The company maintains that firms under CCP control can be legally compelled to share data or assist intelligence agencies, making any commercial access an “unavoidable vulnerability.” |
Strategic Standoff: Anthropic vs. The Pentagon
Interestingly, this move to block the CCP comes as Anthropic is in a fierce legal and political battle with its own government.
While Anthropic is cutting off China to protect “national security,” it is simultaneously refusing the U.S. Pentagon’s demand to remove safeguards for fully autonomous weapons and mass domestic surveillance. Amodei pointed out the irony that the Pentagon is threatening to label Anthropic a “supply chain risk”—a label usually reserved for Chinese companies—even as the firm sacrifices millions in revenue to fight Chinese espionage.
“We chose to forgo several hundred million dollars in revenue to cut off the use of Claude by firms linked to the CCP… regardless of the pressure, we cannot in good conscience remove the safeguards the Department of War is requesting.” — Dario Amodei
Impact on Global Tech
The ban is already causing ripples in the international developer community:
- Singapore & Beyond: Tools like ByteDance’s “Trae” code editor (based in Singapore), which rely on Claude’s API, are seeing users scramble for alternatives or refunds due to the new “50% ownership” rule.
- Enhanced Detection: Anthropic has deployed new “Behavioral Fingerprinting” systems to identify API traffic patterns that look like model distillation rather than human use.
