Key Takeaways
The Pentagon declared Anthropic a "supply chain risk" and moved to bar it from federal contracts after the company refused to let its AI power autonomous weapons or conduct mass surveillance.
OpenAI faced the same red-line decision and answered it differently — keeping the same no-autonomous-weapons, no-mass-surveillance restrictions, but building in an operational kill switch rather than walking away from the contract.
President Trump weighed in directly on Truth Social, effectively setting policy by declaring Anthropic off-limits not just at the Pentagon but across the federal government.
Republican-led states, not only blue states, are moving on AI regulation — Axios reported the White House directly intervened to kill a state AI law in Utah.
AI is shaping up as a real 2026 midterm issue, with AI industry money flowing into races the way crypto money did before it.
Washington has always argued with Silicon Valley. What's happening right now is a different kind of fight. The Pentagon — the federal government's single largest technology buyer — declared an AI company a supply chain risk and tried to cut it off from federal work entirely, not because the technology failed, but because the company refused to let its AI power autonomous weapons or run mass surveillance on Americans. My guest has been inside that story since it broke. Maria Curi covers AI and government policy for Axios, co-writing the newsletter AI+ that most of tech policy Washington reads every week, and she walked me through what the Anthropic-Pentagon dispute actually reveals about who gets to set the rules for artificial intelligence in this country.
What Happened Between Anthropic and the Pentagon?
The short version: Anthropic drew a line. "They won't let their AI run autonomous weapons or surveil Americans at scale," Curi told me — a position that, by her reporting, was not well received inside an administration that already viewed the company as, in her sources' words, "a liberal company... a woke company, leftist, radical leftist company." The dispute became public as all four major AI labs — Google, OpenAI, xAI and Anthropic — moved from doing non-classified work with the Pentagon to negotiating access for classified use, under a Pentagon standard requiring the AI be usable "for all lawful purposes." Anthropic read that standard as dangerously broad. "There are no laws today that are contemplating AI and the way that AI can be used in the context of war," Curi explained, paraphrasing the company's own reasoning. "The legal system has not caught up to it, and that's why they felt they needed to put these red lines in the sand."
It didn't stay a contract dispute for long. President Trump posted on Truth Social, according to Curi's reporting, effectively declaring Anthropic a "radical left company" and stating that not just the Pentagon but the entire federal government would stop doing business with it. That's the moment this stopped being a procurement disagreement and became policy, set outside of Congress, outside of any formal rulemaking — on social media.
How Did OpenAI Respond Differently Than Anthropic?
Every major lab faced the same choice. xAI, run by Elon Musk, was first to sign, agreeing to the Pentagon's "all lawful purposes" language — though Curi noted xAI's technology wasn't seen as advanced enough to actually replace Anthropic's Claude on classified work. Google, by her account, stayed almost entirely silent, with the closest thing to a public position being a tweet from chief scientist Jeff Dean expressing the same concerns Anthropic had raised.
OpenAI took a third path, and it's the most instructive one. As Anthropic's own negotiations were falling apart, OpenAI moved quickly to negotiate its own Pentagon contract — CEO Sam Altman has said he wants those contract terms extended to every other lab, though other companies dispute that OpenAI's terms are actually the strongest. What OpenAI landed on, per Curi's reporting: the same red lines Anthropic drew — no autonomous weapons, no mass surveillance — but instead of stating them as conditions, OpenAI built them in as safeguards it controls directly. "They are going to remain operationally in control of the model on the back end, so that if there is ever a situation where that is happening, they could turn off the switch," Curi said. Same restriction, different posture — supervision built into the product instead of asserted as a precondition.
Is the White House Setting AI Policy Through Contracts Instead of Congress?
This is the question Curi keeps circling back to in her own reporting, and it's bigger than one contract. "Is the Pentagon doing more to regulate AI essentially than Congress?" she asked, rhetorically, while walking me through the internal chain of command on this specific dispute. The public back-and-forth ran through Pentagon undersecretary Emil Michael and Anthropic CEO Dario Amodei before escalating to Defense Secretary Pete Hegseth and, ultimately, to Trump himself on Truth Social.
Curi's read on where actual influence sits inside the administration is worth sitting with: David Sacks, the person she describes as closest to Trump's ear on tech policy throughout the White House's AI strategy, has since moved to an outside advisory role — PCAST, alongside Jensen Huang and Larry Ellison — "no longer in those day-to-day meetings." He still matters, in her sources' telling, but he's a step removed at exactly the moment this dispute has escalated to the president's own social media feed. Meanwhile, she said, people close to the administration are telling her privately that a hands-off posture toward AI "may not be as popular anymore" — that not regulating AI may not be what voters in an election year actually want.
Which States Are Leading on AI Regulation?
Curi pointed me somewhere I didn't expect: Republican states, not blue ones. "This is not — or I should say, this is an issue that does not cut neatly in partisan lines," she said. Her example was Utah, where Axios was first to report that the White House directly intervened to kill a state AI law before it could move forward — a preview, in her view, of federal officials stepping in on other states next. She named Utah Governor Spencer Cox specifically as someone worth watching, and argued that plenty of Republican leaders and voters want real guardrails around this technology, even if that instinct doesn't show up cleanly in national party positioning yet.
How Is AI Becoming a 2026 Midterm Election Issue?
Curi's broader point is that AI is colliding with several currents at once heading into the midterms: anti-billionaire sentiment, distrust built up over a decade of social media harm (she pointed specifically to a California court decision finding Meta and YouTube liable for harm to young people), and now serious AI industry money moving into campaigns the way crypto PAC money did before it. Candidates, in both parties, are being forced to choose between that money and what she sees as "increasingly strong voter sentiment" around AI's effect on jobs, kids, data centers landing in people's towns and cybersecurity. "This time around, it's on a lot of voters' minds," she said — a genuinely new dynamic for a beat that, in her words, "has never been a real salient issue in an election" before now.
Find Maria Curi: Maria Curi covers AI and government policy for Axios, where she co-writes the AI+ newsletter with colleague Ashley Gold. Her reporting on the Anthropic-Pentagon dispute and the broader fight over AI regulation runs throughout the week on Axios, not just in the Friday edition.
The Friday Reporter is hosted by Lisa Camooso Miller. Listen wherever you get your podcasts — and subscribe so you never miss a conversation.