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

What It Covers

Under Secretary Emil Michael joins the All-In hosts to cover Operation Epic Fury against Iran, the Pentagon's termination of Anthropic's $200M contract over AI usage restrictions, drone warfare evolution, the strategic China leverage angle behind Middle East operations, and the defense industrial base modernization effort underway at the Department of Defense.

Key Questions Answered

• •AI Vendor Lock-in Risk: Organizations embedding a single AI provider face existential operational risk. Anthropic held the "control plane" over model weights hosted in AWS GovCloud, meaning they could alter or disable outputs at any moment. The Pentagon's lesson: require at minimum two redundant AI providers with identical "all lawful use" terms, and never allow a vendor to hold unilateral shutdown authority over mission-critical systems.
• •Autonomous Weapons Threshold Framework: AI autonomy in weapons systems should be calibrated to consequence severity. Low-consequence scenarios — laser intercepts of incoming drones in open airspace — warrant higher AI autonomy now. High-consequence scenarios in dense civilian areas require human oversight. Golden Dome missile defense represents the clearest current case for full autonomy: hypersonic missiles allow roughly 90 seconds before warhead separation, exceeding human reaction capacity.
• •Iran-China Oil Leverage Strategy: Roughly 19% of China's GDP depends on imported oil, with approximately one-fifth of that supply coming from Iran and Venezuela combined. By disrupting both supply chains simultaneously, the administration enters April's US-China negotiations with significant economic leverage. Adding Russia to potential deal terms would affect roughly 40% of China's total oil supply, creating conditions for a potential grand bargain on trade, Taiwan, and semiconductor access.
• •Defense Procurement Reform — Problem-First Contracting: The Pentagon is shifting from specification-heavy requirements (which produced unbuildable, cost-plus contracts) to operational problem statements. Instead of listing 40 technical specs for a missile, the department now states the operational need — range, payload, effect — and solicits solutions. Startups willing to accept fixed-price contracts with delivery incentives gain a structural advantage over legacy primes under this model.
• •Drone Arsenal Economics: One-way attack drones capable of traveling 500–700 miles with significant warheads currently cost $50,000–$80,000 per unit depending on payload configuration. Future drone warfare will involve AI-controlled heterogeneous swarms — mixed drone types that must deconflict targeting autonomously. China already mandates interoperability across drone manufacturers, giving them a current advantage in swarm coordination that the US is actively working to close.
• •AI Model Commoditization Timeline: The performance gap between leading AI models is narrowing measurably, with confidence intervals on benchmark outperformance shrinking across providers. Within 12 months, model capabilities across Grok, Gemini, OpenAI, and Anthropic are expected to converge significantly. For enterprise buyers, this means negotiating leverage shifts toward price competition. The strategic move now is establishing multi-model infrastructure before commoditization arrives, rather than optimizing for any single model's current capability edge.