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
- 0:00 – 2:30
Emergency pod setup: Emil Michael joins as Under Secretary of War (R&E)
The hosts frame the episode as an emergency discussion and introduce Emil Michael, now serving as Under Secretary of War for Research and Engineering. The group sets the tone: an urgent geopolitical situation plus a major Pentagon/AI controversy later in the show.
- •Emergency episode context and quick cast roll-call (Sacks absent)
- •Emil’s background and current role under Secretary Hegseth
- •Light banter sets up a serious national-security conversation
- 2:30 – 6:03
Operation Epic Fury: what success looks like and whether boots-on-ground is coming
Jason summarizes the reported scope of US/Israel strikes and shifting public rationale for the conflict. Emil explains the administration’s stated objectives, expected timeline, and why he believes a prolonged boots-on-ground scenario is not in the cards.
- •Reported leadership decapitations, strikes, and early casualty figures
- •Stated goals: degrade Iran’s ability to fund/arm proxies, missiles, drones, nuclear capability
- •Emil: operation measured in weeks, not months
- •Boots-on-ground: Emil rejects an Iraq/Afghanistan-style protracted occupation
- 6:03 – 12:20
Strategic framing: leveraging Iran/Venezuela to negotiate a grand bargain with China
Friedberg and Chamath argue the Iran conflict is part of a larger chessboard move aimed at China ahead of high-stakes negotiations. They connect oil dependency, sanctions leverage, and global stability (including Russia) to US bargaining power and deterrence against a Taiwan invasion.
- •Friedberg: Iran oil flows heavily to China; disrupting supply creates negotiating leverage
- •Chamath: China’s slowing growth increases risk of external conflict (Taiwan) as a pressure valve
- •Oil supply math: Iran/Venezuela (and potentially Russia) as key inputs into China’s economy
- •Thesis: leverage now could prevent a larger US–China conflict later
- 12:20 – 16:06
Emil’s rebuttal and the ‘new’ American warfighting model: precision, planning, and limited missions
Emil pushes back on the idea that the administration has launched ‘eight conflicts’ and argues recent actions are limited and distinct. He credits deep operational planning, improved joint capabilities, and accumulated lessons from the GWOT for fast, high-success missions like the Maduro raid and planned Iran contingencies.
- •Emil distinguishes inherited conflicts vs. limited new operations (Houthis, Venezuela, Iran)
- •Operations planned for years (‘on the shelf’) and continuously updated
- •US military experience and joint effects (space/air/sea/cyber) as decisive advantage
- •Venezuela raid cited as example of speed, surprise, and minimal casualties
- 16:06 – 18:05
Rules of engagement reset: from restrictive parity to commander judgment and overwhelming force
Emil argues that technology alone didn’t change outcomes—rules of engagement did. The conversation contrasts prior constraints (legalistic, parity-style limitations) with a shift toward clearer objectives, red lines, and field-level judgment reminiscent of the Powell Doctrine.
- •Emil criticizes prior ROE as operationally dangerous and overly legalistic
- •Shift toward empowering commanders within defined constraints
- •Goal: “unfair fight” to protect US forces and achieve objectives quickly
- •Overwhelming force + clear objectives + exit plan emphasized
- 18:05 – 20:34
Drone dominance and LUCAS: cheap, attritable systems and a ‘DARPA challenge’ procurement style
The group dives into drones as the new center of gravity in warfare and Emil outlines the Department’s push for an ‘arsenal of drones.’ Emil describes the LUCAS concept—long-range, one-way attack drones—alongside a faster, more competitive pathway for vendors to get hardware deployed.
- •Drones as a dominant casualty driver in Ukraine/Russia and a template for future wars
- •LUCAS: long-range, one-way drones (hundreds of miles) at tens-of-thousands cost scale
- •Broader uses beyond combat: base defense, border interdiction, major event security
- •A faster competition model to bring startup-built systems into the force
- 20:34 – 28:39
AI at the edge: swarms, target recognition, comms-denied ‘kill boxes,’ and safety tradeoffs
Emil discusses how AI will increasingly coordinate drone swarms and operate in jammed environments, while acknowledging reliability limits and escalation risks. The chapter explores where autonomy is acceptable (e.g., intercepts) versus where higher human oversight remains necessary (dense civilian areas).
- •Future: heterogeneous drone swarms coordinated partially by AI
- •Edge AI for target discrimination, decoys, and comms-denied operation
- •Reliability: no clear ‘FSD moment’ yet; risk tolerance varies by scenario
- •Golden Dome example: time-critical intercepts may require AI decisioning
- 28:39 – 37:18
Israel’s role and influence debate: Mossad intel, alliance value, and Iron Beam/laser defenses
Jason raises the internal US political debate about whether Israel drives US Middle East policy. The panel argues the US acts in its own interest but benefits from Israeli capabilities and intelligence; Emil addresses the state of Israeli laser defenses and joint tech collaboration.
- •Debate: ‘captured by Israel’ vs. alliance-driven coordination
- •Argument: Israeli intelligence/capability accelerates operational success
- •Iron Beam and directed energy systems: early but real progress
- •Multi-layer defense: space-based + ground-based interception needs
- 37:18 – 41:19
Economic shock channels: Strait of Hormuz disruption and Trump’s maritime insurance backstop
Oil prices and shipping disruptions become the focus, with Friedberg explaining why insurance markets are pivotal to maritime trade. He frames the US stepping in as an insurer as both an economic stabilization tool and a potential opportunity to onshore a major underwriting industry.
- •Hormuz shipping slowdown and war-risk insurance cancellations
- •Friedberg history: Lloyd’s of London and modern reinsurance origins
- •Premium spikes and leverage constraints on shippers (debt requires insurance)
- •US backstop via Development Finance Corporation as stabilization + industry creation
- 41:19 – 42:40
Breaking news: Pentagon designates Anthropic a supply-chain risk—what triggered it
Jason reports Bloomberg’s story and tees up the dispute: autonomy concerns and surveillance fears. Emil responds with his version of events, emphasizing contract terms that restricted core DoD use cases and a growing fear that Anthropic could disrupt operations at critical moments.
- •Anthropic contract cancelled and supply-chain risk designation announced
- •Public framing: ‘murder bots’ and mass surveillance vs. DoD’s lawful-use stance
- •Emil: existing contract terms barred kinetic planning and other mission-critical tasks
- •Core worry: refusal/shutoff/guardrails activating during real-world operations
- 42:40 – 48:45
Inside the Anthropic contract fight: exceptions, control plane risk, and the Maduro-raids question
Emil details months of negotiations where Anthropic offered narrow exceptions rather than a broad lawful-use standard. The turning point: Anthropic allegedly probing whether its software was used in the Maduro operation, raising fears of enforcement actions, outages, or model manipulation.
- •Emil reviews ‘winner’ status from prior AI EO era and Anthropic’s deep gov embedding
- •Negotiations devolved into scenario-by-scenario exceptions vs. future-proof terms
- •Technical architecture: Claude hosted in AWS GovCloud but Anthropic retains control plane/refresh ability
- •‘Was it used in the raid?’ inquiry sparks alarm about operational dependency and insider threat
- 48:45 – 57:24
Industry implications: model ‘deplatforming,’ multi-model strategy, and who’s aligned with DoD terms
The panel generalizes the Anthropic episode into a broader risk: vendors imposing shifting moral/political constraints on critical infrastructure. Emil contrasts Grok, Google, and OpenAI positions, arguing DoD must maintain redundancy, avoid ideology in outputs, and drive competition as models converge.
- •Chamath: AI terms-of-service enforcement resembles deplatforming—now higher stakes
- •Emil: Grok supports all lawful uses; Google supports lawful use (unclass) and is moving toward classified
- •OpenAI portrayed as cooperative and ‘patriotic’ in helping provide redundancy
- •Strategy: multi-model governance now; price competition if/when model performance asymptotes
- 57:24 – 1:22:58
Why make it official? Supply-chain risk designation, valuation talk, and the defense supply chain agenda
Emil argues the supply-chain label isn’t punitive; it’s meant to prevent Anthropic from becoming embedded across the broader defense industrial base where bias or disruption could compound risk. The conversation then pivots to the physical supply chain—critical minerals, batteries, munitions, and scaling defense-tech startups through faster contracting and DARPA pathways.
- •Emil: designation blocks defense contractors from depending on a potentially unreliable/bias-prone model
- •Public spat dynamics: claims of pay-for-play dismissed; dispute framed as operational reliability
- •Defense industrial priorities: critical minerals, batteries, domestic production using Strategic Capital office
- •Defense-tech boom: need faster ‘big wins’ for startups; move away from cost-plus and bureaucratic requirements
- •DARPA highlights: bio-based mineral extraction, AI-era cyber threats; China buildup acknowledged but US retains key advantages
