Trump's October 2025 order restricted Anthropic's Mythos and Fable model families from export on national-security grounds. That control lifted this week, and the mechanism matters more than the headline: Wired reports Anthropic got back into the administration's good graces not by lobbying, but by adding a new safety-testing gate before Claude ships. Ars Technica confirms the sequence: the security measure came first, the global release of Anthropic's models followed, and Fable 5, Anthropic's next flagship, is now greenlit to return after being pulled from that same order. Compare that to OpenAI's move the same week: a 5% equity stake offered to a US sovereign wealth fund, which Ars Technica reports Senator Sanders wanted pushed far higher. One company traded equity for political cover. The other traded a testing protocol for an export license. Only one of those is a compliance framework a regulator can point to next time.

That distinction sets the terms for Anthropic's next hardware move. TechCrunch reports Anthropic is in talks with Samsung for a custom AI chip, the kind of deal that only works if the finished part can leave South Korea and reach Anthropic's training clusters without tripping the same export-control regime that just got resolved. A chip design win in Suzhou or Hsinchu fabrication capacity means nothing if Washington re-tightens the licensing terms Anthropic just cleared. Microsoft, meanwhile, is hedging differently: its new $2.5 billion AI deployment firm, announced the same week, is a bet on distribution infrastructure rather than a custom chip supply chain exposed to export review. Anthropic just proved the safety-testing gate is the price of a lifted export control. Whether that price holds through a Samsung chip contract is the next test, and it lands whenever that deal is finalized and filed with Commerce.