Inside the Iran War and the Pentagon’s Feud with Anthropic with Under Secretary of War Emil Michael

Summary

This emergency All-In episode interviews Emil Michael about the U.S.-Israel operation against Iran, AI-coordinated drone warfare, missile defense, the Anthropic dispute, and defense-industrial modernization. Its strongest contribution is a first-person Pentagon-side explanation of why acceptable-use restrictions and vendor control over a deployed model were treated as operational risks, while the episode’s war analysis remains openly administration-aligned and does not include Anthropic’s response.

The source connects Defense AI Procurement to Defense AI Control Plane Risk: Michael says a model embedded through Amazon Web Services GovCloud and Palantir remained refreshable by its vendor, so model policy, technical control, and continuity could not be separated. It also qualifies Human-Authorized Autonomous Weapons by arguing that acceptable AI risk depends on mission, civilian exposure, and decision time rather than one universal autonomy rule.

Key Claims

• Michael says the Iran operation sought to degrade nuclear, missile, and proxy-support capacity without becoming an Iraq- or Afghanistan-style ground occupation.
• Chamath Palihapitiya and David Friedberg interpret pressure on Iranian and Venezuelan oil flows as possible leverage in a wider U.S.-China negotiation; Michael calls any China benefit second-order.
• Michael says drones caused about 70% of Russia-Ukraine casualties and argues that AI-coordinated, heterogeneous swarms will exceed the scale of human-only control.
• He distinguishes lower-risk defensive decisions from autonomy around civilians and says very short missile-defense timelines may require AI-assisted or automated action.
• Michael says Anthropic’s contract terms restricted kinetic-strike planning and satellite movement, while the department wanted one standard covering all lawful uses rather than scenario-by-scenario exceptions.
• He says Anthropic retained the model control plane for a model served through AWS GovCloud and Palantir, creating concern that the vendor could refresh or change a capability embedded in sensitive workflows.
• Michael characterizes the supply-chain-risk designation as continuity and policy-dependence management rather than punishment; this is the department-side account and is not independently established by the supplied document.
• The episode says alternative providers including xAI, Google, and OpenAI were being pursued to create redundancy and price competition.
• Michael describes a modernization agenda built around drones, attritable systems, procurement reform, new defense primes, critical-mineral and battery supply chains, and strategic lending.
• He presents DARPA as a long-range research engine for cyber defense and biological routes to critical-mineral extraction or synthesis.

Key Quotes

“all lawful use” - Michael’s preferred procurement standard as recorded in the supplied episode summary.

“drone-on-drone and robot-on-robot warfare” - Michael’s description of the direction of battlefield autonomy.

“If a company does not want its tools used for Department of War work, it should not sell to the Department of War.” - Michael’s vendor-policy position as recorded in the supplied summary.

Connections

• All-In, Emil Michael, Jason Calacanis, Chamath Palihapitiya, and David Friedberg - show, guest, and host context.
• Iran, Israel, China, United States, Strait of Hormuz, and Iran Horizontal Escalation / 伊朗横向升级 - war, energy, shipping, and bargaining context.
• Low-Cost Drone Warfare, Drone Defense Economics, and Human-Authorized Autonomous Weapons - drone mass, defensive cost, and autonomy-threshold branch.
• Anthropic, US Department of Defense, Palantir, Amazon Web Services, OpenAI, Google, xAI, and Grok - defense-model procurement and substitution context.
• Defense AI Procurement, Defense AI Supply Chain Risk, Frontier Model Use Policy Conflict, and Defense AI Control Plane Risk - main AI procurement concepts extended or added by the source.
• Anduril, Saronic, Defense Tech Startup Procurement, and Defense Industrial Base Reindustrialization - new-prime, attritable-system, and industrial-capacity branch.
• DARPA, Critical Minerals Geopolitics, Supply Chain Sovereignty, and AI Cyber-Defense Utility - research and strategic-input branch.

Contradictions

• No settled contradiction is adopted. The episode is the Pentagon-side account of a disputed negotiation and does not include Anthropic or the underlying contract, designation notice, classified-use documentation, or technical control records.
• Earlier wiki sources summarize Anthropic’s red lines as mass surveillance and fully autonomous weapons. Michael instead describes bulk public-information collection, kinetic-strike planning, satellite movement, scenario exceptions, and vendor control of model refreshes. These accounts may describe different contract provisions or incompatible framing; the supplied evidence does not resolve which scope is complete.
• The source speaks as though the $200 million contract was canceled and the supply-chain-risk designation was formal, while a same-day Marketplace source said Anthropic reportedly had not yet received the designation in writing and a later source described a broader federal ban. Timing, legal scope, and implementation remain unresolved.