At the Pentagon, OpenAI is In and Anthropic Is Out

Key Takeaways Copied to clipboard!

• The core conflict revolves around whether the Pentagon's demand for an "all lawful use" standard for AI models is substantively different from Anthropic's insistence on explicit prohibitions against mass domestic surveillance and autonomous weapons, leading to Anthropic's designation as a supply chain risk while OpenAI secured a deal with similar stated safeguards.  Copied to clipboard!
• The punitive action taken against Anthropic, including the supply chain risk designation, is described as the most severe punitive action the U.S. government has taken against a major American company this century, potentially chilling future tech industry interactions with the government.  Copied to clipboard!
• Employee activism, including open letters expressing solidarity with Anthropic's stance against military use cases, is noted as a potentially meaningful factor influencing leadership decisions at major AI labs like OpenAI and Google DeepMind.  Copied to clipboard!

Segments Expand All Collapse All

Setting the Stage: AI Dispute

Copied to clipboard!

(00:00:48)

• Key Takeaway: The AI industry experienced a chaotic 48 hours involving a Pentagon dispute with Anthropic and a subsequent deal struck by OpenAI.
• Summary: The episode opens by framing the recent intense developments concerning Anthropic and the Pentagon, followed by OpenAI’s eleventh-hour agreement with the Defense Department. Hosts disclose personal connections to the situation, noting one host is engaged to an Anthropic employee. The central confusion is whether substantive differences exist between the rejected Anthropic deal and the signed OpenAI agreement.

Anthropic’s Red Lines Crisis

Copied to clipboard!

(00:02:24)

• Key Takeaway: Anthropic reached a crisis point with the Pentagon by refusing to compromise on red lines against mass domestic surveillance and autonomous weapons.
• Summary: Anthropic publicly stated it would not compromise on its two red lines, leading the Pentagon to threaten to declare the company a supply chain risk. CEO Dario Amadei invoked conscience in his statement, a rare occurrence in tech leadership discussions regarding government contracts. Discussions continued between the parties even after Amadei’s public statement, attempting to resolve the impasse.

Trump and Hegseth Escalation

Copied to clipboard!

(00:05:38)

• Key Takeaway: President Trump ordered a federal agency halt on Anthropic use, immediately followed by Defense Secretary Hegseth designating Anthropic a supply chain risk, a severe measure usually reserved for foreign entities.
• Summary: President Trump posted a statement directing federal agencies to cease using Anthropic’s technology over six months, avoiding mention of the supply chain risk designation. Shortly after, Defense Secretary Pete Hegseth announced the immediate designation of Anthropic as a supply chain risk, effectively banning contractors from dealing with the company. This designation is significantly stricter than simply losing government contracts and has not typically been applied to major American companies.

OpenAI’s Competing Agreement

Copied to clipboard!

(00:08:36)

• Key Takeaway: OpenAI announced a deal to deploy its models on the Pentagon’s classified network, claiming the agreement incorporates the same safeguards against mass domestic surveillance and autonomous weapons that Anthropic insisted upon.
• Summary: OpenAI CEO Sam Altman previously signaled solidarity with Anthropic’s stance against specific military uses in a leaked message to employees. However, Altman later announced OpenAI’s agreement with the Pentagon, asserting that the military accepted their principles regarding mass domestic surveillance and autonomous weapons. This creates a paradox where two companies claim identical red lines, yet one is penalized while the other secures a deal.

Nuance of Contractual Language

Copied to clipboard!

(00:11:03)

• Key Takeaway: The conflict likely hinges on the interpretation of the ‘all lawful use’ standard, as Anthropic claims the Pentagon offered them similar concessions accompanied by ineffective legalese.
• Summary: The difference between the agreements may lie in the contractual language, specifically regarding the ‘all lawful use’ standard the Pentagon sought. Anthropic suggests the Pentagon’s proposed concessions were undermined by legalese that would render them ineffective, contrasting with OpenAI’s claim that the Pentagon agreed to uphold the principles in law and policy. The legality of current data gathering practices, such as federal agencies buying data from brokers, functionally enables surveillance even if explicitly prohibited by name.

Political Vendetta vs. Substantive Differences

Copied to clipboard!

(00:15:38)

• Key Takeaway: The situation presents two possibilities: either the Pentagon has a political vendetta against Anthropic, or OpenAI conceded to substantive terms Anthropic refused, masked by ambiguous contract language.
• Summary: One possibility is that the dispute is purely political, fueled by ideological clashes and personal animosity between officials like Pentagon Undersecretary Emil Michael and Anthropic’s Dario Amadei. The alternative is that OpenAI agreed to terms that allow the Pentagon to proceed with surveillance or weaponization under the guise of legal technicalities, which Anthropic claims it was offered and rejected. This action is viewed by some former Trump administration members as an attempt at ‘corporate murder’ based on ideology.

Chilling Effect and Employee Activism

Copied to clipboard!

(00:18:00)

• Key Takeaway: The punitive nature of the action against Anthropic sets a chilling precedent for Silicon Valley compliance, while employee activism shows internal resistance to military AI contracts.
• Summary: The severity of the action against Anthropic is unprecedented for a major American company this century, potentially forcing tech companies to align ideologically with the administration to avoid being crushed. Employee activism, including an open letter signed by workers at OpenAI and Google DeepMind, expresses solidarity with Anthropic’s ethical stance. OpenAI is attempting to reassure employees by citing a ‘safety stack’ built into their models, though skepticism remains regarding the effectiveness of such guardrails.

Regulatory Capture and Future Stakes

Copied to clipboard!

(00:26:32)

• Key Takeaway: OpenAI’s success in securing the deal despite the conflict is framed as textbook regulatory capture, highlighting the high-stakes debate over who controls powerful technology: builders or governments.
• Summary: Anthropic’s prior warnings about AI risks are sometimes viewed as a pretext for regulatory capture, but OpenAI’s maneuvering into the dispute is cited as a textbook example of regulatory capture itself. This conflict represents the fundamental question of whether technology builders or national militaries control the deployment of powerful AI systems. The situation underscores the immediate reality of powerful AI systems being rolled out under the vague ‘all lawful use’ standard due to a lack of comprehensive regulation.

Unresolved Questions and Aftermath

Copied to clipboard!

(00:28:16)

• Key Takeaway: Key unresolved issues include the formal legal implementation of Anthropic’s supply chain risk designation and the long-term consumer reaction to the companies’ stances.
• Summary: Future focus areas include the legal details of Anthropic’s supply chain risk designation and what it entails for its other government relationships. Consumer reaction is beginning to show, with reports of users switching from ChatGPT to Claude in support of Anthropic, exemplified by pop star Katy Perry’s public endorsement of Claude Pro. Dario Amadei’s long-term vision, informed by reading ‘The Making of the Atomic Bomb,’ anticipated this exact conflict where powerful AI intersects with national security interests.