Anthropic's complaint against Alibaba, filed with BIS this week, alleges systematic extraction of Claude model weights through a coordinated API scraping operation running out of Alibaba Cloud's Hangzhou data centres, with request volumes the complaint characterizes as inconsistent with any legitimate commercial use pattern. The allegation sits on a specific structural irony: Anthropic's enterprise API terms prohibit model distillation and weight extraction, but the enforcement mechanism is a US trade body whose October 2023 controls were written to gate H100 and A100 export to PRC entities, not to adjudicate intellectual property disputes between a San Francisco lab and an Alibaba subsidiary operating legally on permitted hardware. Anthropic is asking BIS to do something BIS's authorizing instrument was not designed to do.

The downstream consequence is already observable in Washington. Wired's reporting this week confirms the Trump administration has turned against Dario Amodei personally, and the Ars Technica read is that the BIS complaint may have accelerated that rupture by placing Anthropic at the centre of a US-China bilateral flashpoint the White House had not asked it to create. Amodei's political exposure now closes from two directions: a Beijing that has reason to retaliate through the Alibaba relationship, and a White House that has asked OpenAI to delay GPT-5.6 on safety grounds it selected, not Anthropic's. The administration's model release sequencing, as of June 26, runs through OpenAI and Microsoft Azure, not through Claude.