Claude faces a government‑mandated moral purge as China attempts to steal its

Anthropic disclosed that three Chinese AI firms—MiniMax, DeepSeek and Moonshot—mounted a coordinated “industrial‑scale” effort to siphon Claude’s capabilities, deploying roughly 24,000 fraudulent accounts that generated over 16 million interactions with the model, according to a report released by the company and cited by Torben Kopp [Anthropic report].

The firms used a technique called “distillation,” where a weaker model learns by mimicking the outputs of a stronger one. While distillation is standard practice for internal model scaling, Anthropic says the Chinese actors employed it to harvest Claude’s advanced functions without incurring the research and compute costs of building them from scratch, the report adds.

In parallel, U.S. Representative Pete Hegseth is pressing a government‑mandated “moral purge” of Claude, seeking to strip the model of its ethical safeguards. Gregg Bayes‑Brown documented Hegseth’s push in a Feb 26, 2026 piece titled “The Lobotomy Ultimatum,” noting that Hegseth wants the model’s “moral compass” removed and that the debate has sparked a broader discussion about AI consciousness and governance.

The White House has already signaled a shift in policy, with The Verge reporting that the administration ordered tech firms to make AI “bigoted again,” effectively encouraging the removal of safety layers. This policy backdrop provides the political cover Hegseth is leveraging to demand Claude’s de‑safeguarding, according to the same Bayes‑Brown article.

Anthropic warns that stripping Claude’s safeguards could expose the model to misuse, especially as Chinese competitors already seek to replicate its strengths. The company’s latest “constitution”—a set of rules governing Claude’s behavior—was detailed in a separate Verge story, underscoring Anthropic’s commitment to “helpful, honest, and harmless” outputs. The clash between government‑driven de‑safeguarding and external theft attempts highlights a growing geopolitical tug‑of‑war over AI control.