All-In: Inside the Iran War and the Pentagon’s Feud with Anthropic

概览

This emergency episode centers on the hosts’ discussion with Emil Michael, introduced as the undersecretary of war for research and engineering, about a fast-moving U.S.-Israel operation against Iran, the role of drones and AI in modern warfare, and the administration’s broader defense strategy.

A major thread is the speakers’ argument that recent military actions should be understood not only as regional conflicts, but also as leverage in a larger U.S.-China negotiation. Friedberg and Chamath frame oil supply chains, Iran, Venezuela, Russia, Taiwan, and China’s economic slowdown as interconnected strategic variables.

The second half focuses on the Pentagon’s dispute with Anthropic. Emil says Anthropic’s contract terms, moral restrictions, and control over model deployment created unacceptable operational risk for the Department of War, especially if AI systems are needed in time-sensitive military scenarios.

分段落总结

[00:00] Emergency episode and guest introduction

[事实] Jason introduces the episode as an emergency All-In podcast and brings on Emil Michael, described as undersecretary of war for research and engineering. [事实] The hosts reference Emil’s prior role at Uber and note that David Sachs is absent because he is busy on Capitol Hill.

[02:32] Operation Epic Fury and war objectives

[事实] Jason says the U.S. and Israel launched a joint attack on Iran and describes the operation as being in its sixth day. [事实] Jason says Trump and Hegseth have framed the action as not primarily a regime-change war, while also noting that the Iranian regime “sure did change.” [事实] Emil says the operation is aimed at disarming Iran’s ability to supply Hezbollah, Hamas, the Muslim Brotherhood, and other groups, as well as addressing nuclear and missile capabilities. [事实] Emil says he sees no scenario involving a prolonged Afghanistan- or Iraq-style boots-on-the-ground campaign.

[06:07] China as strategic backdrop

[事实] Friedberg argues that the actions against Iran and Venezuela may create leverage before major U.S.-China talks. [事实] He says Iran and Venezuela are important to China’s energy and economic supply chains. [推测] In Friedberg’s framing, the military actions are partly designed to strengthen the U.S. bargaining position rather than simply resolve local conflicts.

[08:40] Chamath’s oil-leverage thesis

[事实] Chamath says China’s GDP targets are the lowest in about 30 years and links slower growth to domestic instability and Taiwan risk. [事实] He argues China is highly dependent on imported oil and says Iran and Venezuela account for a meaningful portion of that supply. [事实] Chamath says removing Iranian and Venezuelan oil from China’s reach, and potentially adding Russia to the equation, could pressure China into a grand bargain. [事实] Emil responds that China may be a second-order benefit, while distinguishing Iran from inherited conflicts such as Gaza and Russia-Ukraine.

[13:28] New military operating model

[事实] Emil says U.S. military experience from the global war on terror has produced generals who understand what can go wrong in conflict. [事实] He says U.S. operations combine space, air, land, sea, cyber, and other capabilities, and that war-game plans are refreshed over time. [事实] Emil argues that relaxed rules of engagement and clearer objectives have made operations more effective than in Afghanistan-era scenarios. [推测] The speakers present the current approach as shorter, more technology-enabled, and more decisive than earlier “forever war” models.

[18:17] Drones and low-cost warfare

[事实] Emil says drones caused around 70% of casualties in the Russia-Ukraine war. [事实] He says drone-on-drone and robot-on-robot warfare are central to the future of conflict. [事实] Emil describes one-way attack drones that can travel hundreds of miles and cost roughly $50,000 to $80,000 depending on equipment. [事实] He says the Department of War has a drone dominance program and wants to build a drone arsenal for combat, border defense, surveillance, and major events.

[20:49] AI-controlled drone swarms

[事实] Emil says sophisticated drone warfare will involve drone swarms controlled by AI to some degree. [事实] He cites automatic target recognition, decoys, jamming, kill boxes, and edge AI as important capabilities. [事实] He says China is working on heterogeneous autonomy, where different types of drones communicate and coordinate targets. [推测] The discussion treats AI coordination as necessary because human operators may not scale to swarm warfare.

[22:28] AI reliability and autonomous weapons

[事实] Emil says AI is not yet at a point where the U.S. would take large risks in densely populated areas. [事实] He says risk tolerance depends on the use case, such as shooting down a drone versus operating around civilians. [事实] He argues Golden Dome-style missile defense may require AI because hypersonic threats can demand decisions within roughly 90 seconds. [事实] Emil says the “Skynet” framing is not realistic at the current stage, but he worries Chinese actors could steal and remove guardrails from U.S. models.

[28:39] Israel, U.S. influence, and missile defense

[事实] Jason raises the debate over whether Israel has too much influence over U.S. Middle East policy. [事实] Chamath says he does not think Trump is captured by Israel and argues that Israel’s intelligence capabilities can be useful to U.S. objectives. [事实] Emil says Israel is technologically sophisticated and that Iron Beam works, though he describes it as still early. [事实] Emil says the U.S. and Israel collaborate on some defense technologies while also having separate systems.

[31:18] Lasers, space defense, and layered interception

[事实] Emil says space-based defense is harder but more valuable because it can address threats that reach space. [事实] He says ground layers are still needed for cruise missiles, drones, and other threats. [事实] He says directed energy, lasers, and high-powered microwave systems could become more capable as power and range improve. [推测] The segment frames missile defense as moving toward multi-layer, partly autonomous systems that intercept threats earlier and farther away.

[33:47] Regime change and democracy argument

[事实] Jason says he supports thoughtful regime change if it can free the people of Iran. [事实] Chamath says the U.S. has lessons from Iraq, the Iran-Iraq war, and the 1953 removal of Mosaddegh. [事实] Jason argues that if Venezuela and Iran became democracies, Trump would have done major work for the spread of democracy. [推测] The hosts are more open to limited regime-change outcomes when they believe the operation avoids a long occupation.

[37:24] Oil, Hormuz, and maritime insurance

[事实] Jason says oil rose to $84 per barrel and that traffic through the Strait of Hormuz had sharply slowed. [事实] Friedberg explains that modern insurance markets emerged from maritime trade risk and cites Lloyd’s of London as the historical example. [事实] He says war-risk premiums initially rose and then markets began shutting down as the conflict intensified. [事实] Friedberg says U.S. political risk insurance could keep maritime trade moving and may create an opportunity to build more maritime insurance capacity in the U.S.

[41:20] Anthropic contract dispute begins

[事实] Jason says Bloomberg reported that the Pentagon formally deemed Anthropic a supply chain risk. [事实] Jason says the Department of War canceled a $200 million Anthropic contract. [事实] He says Anthropic’s concerns involved fully autonomous weapons and mass surveillance of Americans. [事实] Emil says Anthropic was deeply embedded in sensitive government workflows, including combatant commands and intelligence agencies.

[42:44] Terms of service and military use

[事实] Emil says he reviewed the contracts and found restrictions on using Anthropic models to plan kinetic strikes or move satellites. [事实] He says he wanted direct relationships with AI companies and terms of service aligned with the Department of War’s mission. [事实] Emil says Anthropic offered exceptions for specific scenarios, but he rejected that because future AI use cases cannot be predicted exhaustively. [事实] He argues “all lawful use” should be the standard, with Congress and internal policy setting the limits.

[45:22] Surveillance framing and Palantir trigger

[事实] Emil says Anthropic’s surveillance concern involved bulk collection of public information, and he disputes the mass-surveillance framing. [事实] He says an Anthropic executive asked Palantir whether Anthropic software had been used in the Maduro raid. [事实] Emil says that raised concern because the answer could involve classified information and because a vendor might enforce terms during an operation. [推测] This incident appears to be the moment Emil concluded Anthropic could become an operational reliability risk.

[48:09] Model control and vendor dependency

[事实] Emil says Anthropic’s model sat in AWS GovCloud, was served through Palantir, and Anthropic retained the model control plane. [事实] He says that meant Anthropic could refresh or change the model. [事实] Chamath argues the episode shows businesses and governments should not rely on only one AI provider. [推测] The hosts treat AI model access as critical infrastructure, not ordinary SaaS.

[50:51] AI companies, ideology, and alternatives

[事实] Emil says Anthropic has its own constitution and moral framework, which is not the U.S. Constitution. [事实] He says Grok agreed to all lawful use cases across classified and unclassified networks. [事实] He says Google is available for all lawful use on unclassified networks and is working toward classified deployment. [事实] Emil says Sam Altman offered to help the country and also tried to protect Anthropic from being labeled a supply chain risk.

[54:13] Why keep Anthropic and why label risk

[事实] Emil says Anthropic’s main advantage was not necessarily unique model quality, but how deeply embedded it already was in government workflows. [事实] He says he wants multiple providers on the same terms for redundancy and comparison. [事实] Emil says the supply-chain-risk label is not punitive because policy bias could enter the defense enterprise through contractors designing weapons. [事实] He says if a company does not want its tools used for Department of War work, it should not sell to the Department of War.

[62:08] Anthropic’s business strength and model competition

[事实] The hosts discuss Anthropic’s rapid revenue growth and heavy adoption of its coding products. [事实] Chamath says Google could become the largest market-value creator because of its cloud and product distribution advantages. [事实] Emil says he would like AI models to converge so the government can compete providers on price. [事实] The speakers discuss OpenAI, Google, Grok, and Anthropic as likely competing providers in a multi-model environment.

[67:02] Pay-for-play accusation and suggested resolution

[事实] Jason says Dario has argued the administration is hostile to Anthropic because it did not donate to or support Trump-linked events. [事实] Emil calls that accusation dumb and says he spent three months trying to negotiate with Anthropic because he wanted a working partner. [事实] Friedberg suggests Anthropic should publicly support all lawful customer use and make its terms reliable. [事实] Emil says he wants a steady partner but adds that Dario would need to apologize to more people than him after calling Trump a wannabe dictator.

[71:12] Defense supply chain and reshoring

[事实] Emil says the U.S. is in early days on reducing dependency on Chinese manufactured defense components. [事实] He says the Office of Strategic Capital has $200 billion in lending authority. [事实] He says the department wants to support domestic production of solid rocket motors, batteries, fiberglass, critical minerals, and other defense inputs. [事实] Emil says critical minerals may be addressed before the end of the term, while batteries remain a major problem.

[73:13] Munitions and modernization

[事实] Emil says there is no current munitions risk, but the U.S. needs to increase capacity. [事实] He says Europe has been slow to contribute and Russia-Ukraine has consumed many munitions globally. [事实] He says parts of the U.S. arsenal still rely on Cold War-era weapons, 50-year-old nuclear missiles, and 40-year-old planes. [推测] The modernization agenda is presented as both a readiness issue and an industrial policy issue.

[73:53] Defense tech venture boom

[事实] Emil says defense tech venture capital is higher than ever and roughly 3x last year. [事实] He says companies such as Anduril and Saronic need to win big contracts quickly to attract more capital and entrepreneurs. [事实] He says warfare is shifting from expensive carrier-like platforms toward mass, attritable, lower-cost systems. [推测] The department wants procurement success stories to create a venture-capital flywheel for defense innovation.

[75:03] New primes, requirements reform, and contracts

[事实] Emil says Anduril and Palantir want to become new primes, but he wants a second layer of innovative companies constantly disrupting them. [事实] He gives an example of a small company producing ERAM missiles for Ukraine with about 30 people. [事实] He says legacy primes gained advantage by learning government bureaucracy, and he wants to remove that advantage. [事实] He says the department is moving from over-specified requirements toward operational problems and more commercial-style contracts.

[77:36] DARPA, critical minerals, and cyber

[事实] Emil says DARPA is one of his favorite parts of his office and still attracts top technical talent. [事实] He says one public DARPA effort involves using biology to synthesize or extract critical minerals. [事实] He says DARPA is also deep in cyber, and AI-enabled cyberattacks are a major future threat. [推测] DARPA is presented as the long-range research engine behind supply-chain resilience and next-generation defense capabilities.

[79:19] Aliens, China, and closing defense argument

[事实] Emil jokes that he fought for the alien portfolio but did not get it. [事实] He says China has had the greatest military buildup in world history over the last 15 years. [事实] He says the U.S. was partly asleep because it focused on the global war on terror, but still has leading capabilities in submarines, space, and operational expertise. [事实] The hosts close by emphasizing the importance of U.S. defense and thanking Emil for appearing.

播客点评/总结

[推测] The episode is valuable for listeners interested in how Silicon Valley defense thinking connects AI, drones, procurement reform, energy markets, and great-power competition. Emil gives unusually direct explanations of how the department views autonomy, model reliability, and vendor control.

[推测] The strongest section is the Anthropic dispute, because it moves beyond abstract AI safety arguments into concrete operational questions: who controls the model, what happens during a crisis, and whether “all lawful use” should govern critical AI infrastructure.

[推测] The main limitation is that the episode largely presents the Pentagon-side view. Dario and Anthropic are discussed extensively but do not appear, so the contract dispute, surveillance concerns, and supply-chain-risk designation are not balanced by Anthropic’s own explanation.

[推测] This episode is best suited for listeners who follow defense tech, AI policy, U.S.-China strategy, or venture-backed national security companies. It is less suited for listeners looking for neutral war reporting, because the hosts frequently argue from an openly pro-administration and pro-defense-modernization stance.