All-In PodcastHow the Pentagon cut Anthropic over terms-of-service clauses
Anthropic's contract barred defense uses. The Pentagon called it a supply chain risk; Emil Michael frames LUCAS drones as the model that replaces it.
CHAPTERS
Emergency pod setup: Emil Michael joins as Under Secretary of Defense for R&E
Jason frames the episode as an emergency update and introduces Emil Michael, describing his background from “Team Uber” to senior Pentagon leadership. The hosts banter briefly before turning to the breaking geopolitical situation.
- •Episode context: rapid-response “emergency” recording
- •Emil’s role: Under Secretary of Defense for Research & Engineering (R&E) reporting into SecDef
- •Light opener/banter with Chamath and Friedberg before hard news
- •Expectation-setting: discussion will focus on war status and defense tech
War with Iran: what happened, what’s the objective, and what “success” means
Jason summarizes the early phase of Operation Epic Fury and competing narratives about why the U.S. is engaged. Emil describes the operation as a “weeks not months” effort focused on degrading Iran’s ability to fund and arm proxy groups and limiting missile/nuclear capabilities.
- •Jason lists reported casualties, leadership decapitation, and escalation signals
- •Administration messaging tension: “not regime change” vs “regime changed” rhetoric
- •Emil’s definition of success: disarm Iran’s ability to supply proxy terror groups and missile/drone capabilities
- •Boots-on-ground question: Emil argues no Iraq/Afghanistan-style occupation scenario
China leverage theory: Iran/Venezuela as bargaining chips before a China summit
Friedberg and Chamath argue the Iran and Venezuela actions primarily build leverage for upcoming negotiations with China. They link China’s oil dependency and slowing growth to Taiwan risk and propose a grand bargain framework.
- •Friedberg: 90% of Iran’s oil goes to China; supply disruption creates negotiation leverage
- •Chamath: China’s lower GDP targets and instability increase Taiwan invasion incentives
- •Oil leverage: Iran + Venezuela (and potentially Russia) are framed as major inputs to China’s energy security
- •Thesis: U.S. aims to prevent a larger U.S.–China conflict by constraining strategic resources
Emil’s pushback: scope of conflicts and the “second-order” China benefit
Emil disputes the framing that the administration has launched many new wars, arguing several were inherited or limited in scope. He concedes China leverage may be a second-order benefit but emphasizes operational goals and coalition security priorities.
- •Emil distinguishes inherited conflicts (Gaza, Russia–Ukraine) from new operations
- •Houthis framed as freedom-of-seas enforcement rather than a “war”
- •Venezuela described as a rapid “one night” raid
- •China leverage acknowledged as possible but not primary driver, in Emil’s view
Why recent operations look “smoother”: planning, training, tech, and rules of engagement
Jason presses Emil on why recent missions (e.g., Venezuela) appear unusually effective with minimal casualties. Emil credits long-planned war games, experienced leadership shaped by GWOT lessons, and relaxed/clarified rules of engagement enabling decisive action.
- •Years of contingency planning and refreshed tactics enable rapid execution
- •Leadership experience: GWOT veterans now senior commanders; Chinese lack comparable combat experience
- •Rules of engagement reform: less restrictive constraints vs prior Afghanistan-era parity rules
- •Doctrine shift: clear objectives, overwhelming force, then exit (Powell-style framing)
Drones and autonomous warfare: swarms, heterogeneous autonomy, and AI at the edge
The conversation turns to drone proliferation and how Ukraine–Russia accelerated drone-on-drone combat. Emil outlines the Pentagon’s view of future warfare: AI-assisted swarms, autonomous target recognition, comms-denied “kill boxes,” and interoperability across diverse platforms.
- •Drone warfare inflection: high share of battlefield casualties driven by drones
- •Future concept: AI-coordinated swarms and “heterogeneous autonomy” across different drone types
- •Edge AI uses: decoys discrimination, automatic target recognition, comms-denied search behaviors
- •China’s advantage: ability to force interoperability across manufacturers and scale swarms quickly
LUCAS and “drone dominance”: building an arsenal and cutting time-to-field
Emil explains the Low-Cost Unmanned Combat Attack Systems (LUCAS) effort and broader “drone dominance” push to scale affordable systems. The hosts compare it to DARPA-style challenges meant to speed procurement and adoption.
- •LUCAS one-way attack drones: long range (hundreds of miles) with relatively low unit costs
- •Goal: create a mass “arsenal” of drones rather than exquisite, scarce platforms
- •Use cases beyond war: base defense, surveillance, border interdiction, event security
- •Procurement intent: faster iteration cycles and wider supplier participation
AI safety vs mission reality: where autonomy is acceptable (Golden Dome example)
Jason questions hallucination risk and the threshold for AI to outperform humans in lethal contexts. Emil argues autonomy acceptance should be scenario-dependent, citing missile defense (Golden Dome) as a time-critical domain where human reaction is insufficient and risk is lower.
- •Emil rejects a single autonomy standard; risk tolerance varies by environment and target type
- •Golden Dome scenario: hypersonic threats and decoys require sub-minute discrimination
- •Near-term focus: basic autonomy and human-tethered systems (e.g., collaborative aircraft)
- •Concern raised: adversaries may steal models and remove guardrails, forcing U.S. to remain capable
Israel’s influence and missile defense tech: Iron Beam/laser systems and collaboration
The hosts address claims that U.S. policy is overly driven by Israel. Discussion shifts to Israeli air defense innovation, directed energy prospects, and how U.S.–Israel technology collaboration works in practice.
- •Debate: whether Israel “captures” U.S. decision-making; panel rejects the claim
- •Israel as a high-capability ally with strong intelligence and rapid tech deployment incentives
- •Directed energy roadmap: ground and space layers for drones, rockets, and missiles
- •Shared development: mix of U.S. and Israeli tech and collaborative integration
Economic fallout: Strait of Hormuz risk, oil prices, and U.S. stepping in as insurer
Jason highlights shipping slowdowns, skyrocketing war-risk premiums, and oil price spikes. Friedberg explains the historical role of maritime insurance and argues the U.S. government’s backstop could stabilize trade and even seed a new domestic insurance industry.
- •War-risk insurance market freezes disrupt tanker traffic and global energy supply
- •Friedberg’s primer: Lloyd’s of London origins and modern reinsurance syndicates
- •Premium spike and subsequent coverage pullback increases inflationary pressure
- •U.S. political risk insurance backstop framed as strategic stabilization and industrial opportunity
Pentagon vs Anthropic: contract cancellation, “supply chain risk,” and the terms-of-service fight
Breaking news triggers a deep dive into why Anthropic’s contract was canceled and why it was labeled a supply chain risk. Emil claims Anthropic’s terms prohibited key defense uses and that the company sought to police operational usage, creating mission risk.
- •Emil: discovered restrictive clauses—no kinetic strike planning, satellite operations limits, broad prohibitions
- •Negotiations devolved into exception-by-exception allowances viewed as unworkable
- •Trigger: Anthropic exec allegedly sought classified info about model use in the Maduro raid via Palantir
- •Core Pentagon concern: vendor could shut off/alter model at a critical moment (control plane/weights)
Vendor reliability and deplatforming risk: multi-model strategy, OpenAI/Google/Grok positions
Chamath argues the incident exposes systemic business risk: AI providers can impose shifting ideology through access control. Emil contrasts vendors: Grok supports “all lawful,” Google supports lawful use (unclassified, building toward classified), and OpenAI moves quickly to support redundancy.
- •Lesson: dependency on a single model vendor creates operational and corporate risk
- •Emil: Anthropic hosted in AWS GovCloud but retained control plane and refresh authority
- •Redundancy strategy: maintain multiple models and compete on price as capabilities converge
- •Political framing rejected by Emil; he argues it’s mission reliability, not donations or “pay-to-play”
Defense industrial base and venture opportunity: onshoring, munitions, and procurement reform
The discussion broadens to supply chain vulnerabilities, critical minerals, batteries, and industrial capacity. Emil outlines reforms: shifting from rigid requirements to problem statements, moving toward commercial-style contracting, and enabling startups to win meaningful contracts faster.
- •Supply chain priorities: critical minerals, batteries, and other “critical 20” components
- •Office of Strategic Capital: lending authority used to catalyze domestic capacity
- •Procurement reform: fewer requirements, more outcome-driven solicitations, more fixed-price incentives
- •Defense tech boom: rising VC interest; need “quick wins” to create a flywheel beyond legacy primes
DARPA and future threats: bio-based mineral extraction, cyber, and China’s military buildup
Emil describes DARPA as a continuing engine of frontier innovation and highlights public work on biologically synthesizing critical minerals and advancing cyber defenses. He closes with a sober assessment of China: massive buildup, but U.S. maintains key advantages that must not be allowed to erode.
- •DARPA focus example: biology-driven extraction/refining to leapfrog mineral bottlenecks
- •Cyber as an AI-amplified threat domain requiring continuous innovation
- •China: fastest military buildup in modern history while U.S. focused on GWOT
- •U.S. advantages: operational experience, subs, space capabilities—paired with urgency to avoid complacency
