United States — Sovereign AI

Trump signs EO directing two parallel tracks: (1) hardening US Government + private-sector cyber defenses against advanced AI threats — CNSS, DoW, CISA each tasked with immediate action on national-security, defense, and civilian federal systems respectively; (2) a voluntary benchmarking + review framework for secure development and release of frontier models, including a process for developers to give the government up-to-30-day early access prior to release to other trusted partners.

Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei subjecting Mythos 5 and Fable 5 models to export controls for any location outside the US and to all foreign persons within the US, including foreign-national Anthropic employees. Call-chain detail (Grok dialogue 13 + Axios BM25 pull): Amazon CEO Jassy raised concerns to senior Trump admin officials June 9-12; multiple calls from ~half a dozen senior officials reached Amodei on Friday June 12/13, including Cairncross, Treasury Secretary Bessent, Commerce Secretary Lutnick, Under Secretary Kessler, plus White House staff. Sequence: Jassy -> Cairncross/Bessent/Lutnick -> NSA-provided proof of jailbreak method -> export-control directive issued that evening. Amodei reportedly received LIMITED SPECIFICS on the exact concern. Anthropic complied by disabling both models globally for all customers; other Claude models unaffected. Anthropic position: the bypass found only minor security flaws that other publicly available models can also find.

National Security Presidential Memorandum 11 on Artificial Intelligence in the National Security Enterprise. Rescinds and replaces Biden NSM-25 (2024). Directs military/intelligence/federal departments to accelerate AI adoption across warfighting and intelligence operations. Four-pillar framework: adoption, adaptation, assurance, accountability. Establishes governance guardrails, talent reserve, and framework ensuring AI systems deployed in national security contexts cannot be disabled or altered without federal government approval. Names senior officials including Secretary of War Pete Hegseth and NSA Director Gen. Joshua Rudd to work with private-sector companies on security protocols preventing advanced model theft/copy/compromise.

June 13-15: UK PM Sir Keir Starmer's government lobbied the White House to restore British access to Anthropic Fable 5 + Mythos 5. June 16: senior White House source told Telegraph there was "zero chance" of UK carveout. Administration indicated any eventual rollback would apply globally rather than allied-state-specifically. Senior White House official to NY Post: "We can't have frontier models running amok." CityAM + Telegraph used "kill switch" framing in mainstream finance coverage for first time outside policy-press contexts.

At G7 Evian working-lunch (June 17) on "Ensuring a safe, rapid and effective deployment of artificial intelligence", Anthropic CEO Dario Amodei and Google DeepMind CEO Demis Hassabis called for a US-led AI coalition to shape global rules and standards. Counter-framing to the allied-state diversification narrative: rather than allied states building parallel sovereign capacity (Canada middle-powers, EU CADA, UK Sovereign AI Fund), they proposed a US-anchored multilateral standards architecture. OpenAI's Altman attended but is not on record advocating the same coalition framing.