U.S. Commerce Secretary Howard Lutnick: We Trust Anthropic
U.S. Commerce Secretary Howard Lutnick told Axios during the G20 Innovation Ministers' meeting in North Carolina that Anthropic "did what we asked" and has returned to "the right side," stating, "We trust Anthropic."
Co-founder Tom Brown also spoke on the same day, emphasizing data center and power licensing. The two parties publicly split in February this year: the President ordered federal agencies to immediately stop using the company's technology, and Defense Secretary Pete Hegseth listed it as a national security supply chain risk, restricting defense contractors from dealing with it. The Treasury, Housing Finance, and other agencies followed up by ceasing the use of Claude. The General Services Administration removed it from the government AI platform and procurement catalog.
In June, the Commerce Department restricted access to Mythos 5 and Fable 5 for foreign nationals due to export controls, leading the company to temporarily take both models offline for all customers. Subsequently, Lutnick sent a letter stating that progress had been made in risk management, allowing some customers to restore access to Mythos, while Fable was later reopened to the public, with the company promising future release agreements. On August 27, Judge Rita Lin of the Northern District of California ruled that the Pentagon's designation of supply chain risks was illegal, retaliatory, and lacked legal support. The controversies included domestic mass surveillance and autonomous weapon applications, with the company refusing to comply with military requests.
The company had previously received a two-year, approximately $200 million cutting-edge capability prototype agreement from the Department of Defense in 2025 and claimed that Claude had entered classified networks and national laboratories. This verbal "trust" does not equate to a full restoration of all military contract terms.
From supply cuts and lawsuits to joint appearances, the government has tied model release rights to government procurement. The trust statement is diplomatic language, while the court injunction is another matter.
In market mechanisms, buyers are the federal and defense systems that want cutting-edge models while adhering to usage terms, and sellers are model companies differentiating their products based on security policies. The driving forces are political reconciliation and court rulings, making it event-driven. Funding is being redistributed between government contracts, corporate subscriptions, and export compliance. The beneficiaries are Anthropic, which can reopen some federal and commercial channels, while the pressured parties are those departments that designated supply chain risks as a quick punitive tool, and specific military needs still blocked by usage terms.
Source: Public Information
ABAB AI Insight
Anthropic has written "no surveillance, no autonomous firing" into its product boundaries, clashing with the Pentagon's demand to include all legitimate military uses in contracts. Trump used a government-wide ban as punishment, while Hegseth upgraded the label typically reserved for hostile suppliers. The emergency shutdown triggered by an Amazon paper in June proved that export controls can act as a switch. Brown went to the G20 to mend relations, Lutnick concluded with "trust," and the judge first invalidated the blacklist. Three lines are running parallel: negotiation, litigation, and export licenses.
The capital path is that federal contracts and classified network deployments were once core assets of its national security narrative, with the $200 million prototype agreement serving as a ticket. The ban cut off not only direct orders but also the defense contractor ecosystem. The reconciliation conditions are release agreements and partial model controls, effectively handing part of the product roadmap to the Commerce Department's calendar. The motivation is to survive and retain corporate clients; the strategy is to use the co-founder's public appearance to exchange for verbal vindication while keeping the court's retaliatory designation in place.
Analogous cases include Huawei's supply chain break after being placed on the entity list, TikTok's forced sale negotiations, and military contractors being suspended from awarding contracts due to compliance terms. This is the first time AI companies have countered sovereign procurement terms with "security policies." The phase has shifted from ideal terms confrontation to conditional re-engagement, not yet returning to frictionless supply before February.
This is a regulatory change. The mechanism of change is that model usage rights have been written as a national security switch, which can be activated by presidential posts, ministerial designations, or simultaneous export signals. The court has limited the arbitrariness of designations, but the Commerce Department still controls the release pace of next-generation models. Trust can be verbally restored, but the next release may still face another 90-minute shutdown. Pricing power has shifted from product safety cards to government licensing calendars.
ABAB News · Cognitive Laws
- Once model terms are written as national security switches, contracts will first lose power based on parameters.
- Verbal trust is restored quickly, but procurement catalogs are restored slowly.
- Security policies can serve as selling points but can also act as supply cut orders.