What it is
“Dual-use” is an export-control term for technology that has both civilian and military application — a classification long applied to encryption software, machine tools, and semiconductor manufacturing equipment. Since January 2025, the United States has extended this logic explicitly to AI itself. The Commerce Department’s Bureau of Industry and Security (BIS) issued a rule on 13 January 2025, the “Framework for Artificial Intelligence Diffusion,” that placed advanced AI model weights (models trained above roughly 10^26 computing operations, under a new export classification, ECCN 4E091) and the semiconductor clusters used to train them under a licensing regime modeled on arms-control precedent. The rule sorted the world into three tiers: a low-risk group of roughly 19 “AI Authorization Countries” including Japan, the UK, Canada and EU members, who get streamlined access; a middle tier of unaligned countries; and a high-risk tier of arms-embargoed states and Macau, who face a presumption of denial.
What had been treated as commercial software was thereby reclassified as a strategically controlled technology — the same posture the US takes toward missile components or nuclear technology, not toward ordinary exports. The distinction crossed from theory into practice on 12–13 June 2026, when Commerce Secretary Howard Lutnick ordered Anthropic to suspend global access to its Claude Fable 5 and Claude Mythos 5 models under the Export Administration Regulations (EAR), citing a claimed jailbreak with national-security implications. It was the first time the US government directly compelled an AI company to revoke a deployed model based on the nationality of its users. Because Anthropic had no way to verify user nationality in real time, it could not comply selectively — it disabled both models worldwide, turning a targeted restriction into a global outage. The episode established, in effect if not yet in settled doctrine, that AI software — not only the chips that train it — can be pulled from the market with the same speed and unilateral authority previously reserved for hardware.
Why it matters for AI governance and narratives
The phrase “state-controlled strategic asset” carries opposite valence depending on who deploys it. Japan’s own AI Promotion Act (formally the Act on the Promotion of Research, Development and Utilisation of Artificial Intelligence-Related Technologies), enacted 28 May 2025, names AI a strategic resource for national development as one of its four founding principles — but frames this as an invitation to build and adopt AI, backed only by soft-law guidance rather than binding restriction. The Anthropic suspension inverted that framing: for Japanese developers who had built products atop a foreign frontier model, “strategic asset” status meant their supply chain could be severed overnight by a foreign government’s national-security determination. The same term that Tokyo uses to encourage AI adoption became, from Washington, a designation that demotes a commercial product to an instrument of state control. This is precisely the kind of framing contest the observatory tracks: identical vocabulary recruited by different ecosystems to justify opposite postures — openness versus lockdown, industrial policy versus arms control.
Key facts and dates
- 13 January 2025: BIS issues the AI Diffusion Framework rule, imposing licensing on advanced model weights and AI training chip clusters and establishing the three-tier country system.
- 28 May 2025: Japan enacts its AI Promotion Act, textually designating AI a strategic national resource — as encouragement, not restriction.
- 12–13 June 2026: Commerce orders Anthropic to suspend Claude Fable 5 and Mythos 5 globally under EAR authority, citing a national-security-implicated jailbreak — the first direct US government revocation of a deployed AI model.
- 24 June 2026: Within two weeks, Tokyo-based Sakana AI launches its Fugu model explicitly marketed as offering “frontier capability without the risk of export controls,” while Chinese cybersecurity firm 360 unveils its Tulongfeng model the same day, with founder Zhou Hongyi openly calling vulnerability-finding AI “a national strategic asset” — embracing, rather than resisting, the framing that Japanese and American developers experienced as a demotion.
Coverage of the export-control action itself relies on industry legal analysis and technology press rather than a located Commerce Department press release or Federal Register notice (the Federal Register site blocked automated access during this research), so some operational details of the June 2026 order should be treated as reported rather than verified against primary text.
Where to learn more
- WilmerHale: BIS Issues Long-Awaited Export Controls on AI — law-firm summary of the January 2025 BIS Diffusion Framework rule and its tier system.
- Cloud Security Alliance: Fable 5 Suspension research note — detailed account of the June 2026 Anthropic model suspension and its enforcement mechanics.
- International Bar Association: Japan’s emerging framework for responsible AI — background on Japan’s AI Promotion Act and its “AI as strategic resource” principle.
- TechCrunch: Asian AI startups launch Mythos-like models as Anthropic’s export ban drags on — reporting on Sakana AI’s and 360’s responses, including the contrasting “strategic asset” framing.