Hard Fork
At the Pentagon, OpenAI is In and Anthropic Is Out
This debrief was written and voiced by AI from the episode's audio.
🎧 AI Debrief
🎙️ Full Episode
Summary
This bonus episode of Hard Fork covers a dramatic 48-hour period in which the Pentagon declared Anthropic a 'supply chain risk' after contract negotiations collapsed, while OpenAI simultaneously swooped in to sign a deal with the Pentagon covering the same AI use cases Anthropic had refused to concede on. Hosts Kevin Roose and Casey Newton — the latter of whom is personally entangled as his fiancé works at Anthropic — break down the sequence of events: Anthropic CEO Dario Amodei drew two hard red lines (no mass domestic surveillance, no fully autonomous weapons), the Pentagon pushed back demanding an 'all lawful use' standard, Trump posted on Truth Social ordering all agencies to stop using Anthropic, and then Pete Hegseth escalated further by declaring Anthropic a supply chain risk on X. The central tension of the episode is whether OpenAI actually negotiated meaningfully different protections than what Anthropic was offered, or whether this is purely a political vendetta driven by ideological disdain for Anthropic's leadership. The hosts argue the critical nuance lies in the phrase 'all lawful use' — because the U.S. has no meaningful AI regulation and no national privacy law, what is technically legal (e.g., purchasing data broker data on millions of Americans) is functionally equivalent to mass surveillance. Anthropic wanted explicit contractual prohibitions; OpenAI appears to have accepted assurances grounded in existing law, which may be hollow. The episode frames this as a landmark moment: the most punitive action a U.S. government has taken against a major American company in modern history, with profound implications for who controls powerful AI — its builders or the governments that want to weaponize it.
Key Takeaways
The 'all lawful use' standard is the crux of the entire dispute. The U.S. has no national AI regulation or privacy law, meaning activities that are functionally surveillance — like buying bulk data from data brokers — are technically legal. Anthropic refused to accept this standard; OpenAI appears to have accepted it.
highThe Pentagon's designation of Anthropic as a 'supply chain risk' is, by the hosts' assessment, the most punitive action the U.S. government has taken against a major American company in modern history — a designation previously reserved for Chinese chip suppliers and foreign entities like Kaspersky Lab.
highAnthropic claims it was offered a similar deal to OpenAI's but that it contained 'legalese that would have made the protections ineffective' — meaning OpenAI's deal may not be substantively different, just more palatable in framing.
highThe political dimension is real: Pentagon officials explicitly used ideological language ('radical left woke company') and there is reported personal animosity between Pentagon undersecretary Emil Michael and Dario Amodei, suggesting this isn't purely a contractual dispute.
highOpenAI's proposed 'safety stack' — technical guardrails built into the model to prevent misuse — is viewed skeptically by experts, who argue AI models cannot distinguish whether data fed to them was legally or illegally gathered.
mediumEmployee activism at OpenAI, Google DeepMind, and other AI labs produced an open letter in solidarity with Anthropic's position, signaling that internal pressure from employees remains a meaningful check on corporate military contracting decisions.
mediumSilicon Valley's broader political capitulation to the Trump administration — from Tim Cook's golden trophy to Greg Brockman's $25M PAC donation — is the systemic backdrop that made this confrontation between Anthropic and the Pentagon almost inevitable for any company that didn't align politically.
mediumNotable Quotes
“These threats do not change our position. We cannot in good conscience accede to their request.”
— Dario Amodei (quoted)
Amodei's public statement before the deadline, described by Casey as remarkable because invoking 'conscience' as a reason to defy the government is something no major tech leader has done since Trump's reelection.
“The United States Of America will never allow a radical left woke company to dictate how our great military fights and wins wars.”
— Donald Trump (quoted)
Trump's Truth Social post ordering all federal agencies to cease using Anthropic's technology, revealing the explicitly ideological framing of the dispute.
“Effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic.”
— Pete Hegseth (quoted)
Hegseth's X post escalating beyond Trump's statement to formally designate Anthropic a supply chain risk — a designation with severe commercial consequences.
“That is honestly how the Chinese government regulates its tech companies. Either you get on board with the party or they crush you.”
— Casey Newton
Casey's assessment of the Pentagon's actions against Anthropic, noting that even former Trump administration members called this an 'attempted corporate murder' based on ideology.
“This is the same company that told us it was gonna build safeguards to make sure that Sora couldn't be used to make images of Bryan Cranston. So I'm just gonna suggest that sometimes when OpenAI tells you it's gonna build guardrails, they don't actually show up on time.”
— Casey Newton
Skepticism toward OpenAI's promised 'safety stack' as a meaningful technical protection against military misuse.
Predictions
The truth about whether OpenAI's Pentagon deal actually protects against domestic surveillance and autonomous weapons will eventually come out — either through whistleblowers, leaks, or documented misuse.
— Casey Newton · Confidence: high · Years, not months
AI systems will be used for what is functionally mass domestic surveillance under the cover of 'all lawful use,' because purchasing data broker data on Americans is legal and AI cannot distinguish legally from illegally gathered data.
— Casey Newton · Confidence: medium · Near-term
Autonomous weapons ('murder bots') represent a medium-term risk enabled by this type of unrestricted military AI deal.
— Casey Newton · Confidence: medium · 3-5 years
The Pentagon's punitive treatment of Anthropic will have a significant chilling effect across Silicon Valley, further accelerating political alignment with the Trump administration among tech companies.
— Kevin Roose · Confidence: high · Ongoing
Some consumers will switch from ChatGPT to Claude in protest of OpenAI's Pentagon deal, though the scale of this shift is uncertain.
— Kevin Roose · Confidence: low · Already beginning
Action Items
Scrutinize the actual contract language if/when it becomes public — don't rely on press releases or X posts from either OpenAI or the Pentagon to understand what was actually agreed to.
Both hosts emphasize that the entire story lives in the contractual details, and public statements from all parties have been deliberately vague or ambiguous.
If you work at an AI company with military contracts, read the contracts your employer has signed and understand specifically what use cases are and aren't prohibited — and be prepared to blow the whistle if violations occur.
Casey argues that AI employees are one of the few remaining checks on military misuse of AI, and that internal pressure has historically moved these companies.
Understand that data broker purchases of American personal data are legal and that there is currently no national privacy law — factor this into your threat model when evaluating government AI use case assurances.
The 'all lawful use' standard only means as much as existing law covers, and current law leaves massive surveillance-equivalent activities unregulated.
Follow the formal status of Anthropic's supply chain risk designation closely — as of recording, it exists only as a social media post, not formal legal action, and Anthropic has said it will fight it in court.
The designation's actual legal and commercial consequences remain unclear and could change significantly depending on whether it is formalized.
Read 'The Making of the Atomic Bomb' by Richard Rhodes to understand the historical framework Dario Amodei has used to prepare Anthropic for exactly this kind of government confrontation over transformative technology.
Kevin and Casey note that Amodei bought this book for all Anthropic employees specifically because he anticipated this type of conflict between AI labs and governments.