Pentagon Threatens to End Anthropic Work in Feud Over AI Terms

Summary

Balance of Power Podcast — Detailed Episode Summary

Episode Title: Pentagon Threatens to End Anthropic Work in Feud Over AI Terms
Date: February 24, 2026
Host(s): Joe Mathieu & Kailey Leinz, Bloomberg Washington Correspondents
Featured Guest(s): Mike Shepard (Bloomberg Senior Editor for Technology), Gregory Allen (Senior Advisor at CSIS, former DoD AI official), Rick Davis (Republican strategist), Jeannie Shan Zaino (Democratic analyst)
Main Theme:
A deeply reported exploration of the escalating feud between the Pentagon and leading AI lab Anthropic over the ethical, legal, and practical guardrails for AI deployments in national security, and the broader implications for AI policy, geopolitics, labor, and economic transformation as highlighted on the eve of a consequential State of the Union.

---

1. Episode Overview

This episode dives into a breaking standoff between the U.S. Department of Defense (DoD) and Anthropic, a major AI model provider, as the Pentagon threatens to end Anthropic’s contract unless they lift restrictions on AI use for potential military applications like autonomous weapons and domestic surveillance. The discussion further branches into the rapidly evolving landscape of defense-focused AI, government/corporate tensions, and the looming societal effects of advanced AI systems — set against the backdrop of the upcoming State of the Union and growing global instability.

---

2. Key Discussion Points & Insights

A. Anthropic’s Pentagon Feud: Setting the Stage

• Background: Anthropic (maker of Claude) is the primary AI model used for DoD classified systems. Recently, rivals like Grok (XAI) and Google’s Gemini have been cleared for similar use, intensifying industry competition.
• The Core Dispute:
  • Anthropic insists on its models not being used for “domestic mass surveillance” or “human-free weapons.”
  • Pentagon (Sec. Pete Hegseth) is demanding “unfettered access” for any lawful military use, threatening harsh penalties including supply chain blacklisting and potentially invoking the Defense Production Act.
• Notable Quote (Mike Shepard, 04:07):
  "The company has said that it is engaged in productive and good faith conversations with defense officials. And yet the tone from the Defense Department has been a little bit different. ... They've made clear that they don't want companies to go in there not being willing to support all war fighting efforts. ... There is clearly a bit of friction and a bit of difference over where the boundaries should be set."

B. AI Policy, Precedent, and the Expanding Field

• Precedent/Competition: With Grok cleared for classified use, the Pentagon has more leverage, but the Anthropic conflict could set significant policy markers for tech/government deals.
• AI Policy Ambiguity: Pentagon wants “legal compliance only” from AI vendors, resisting company-imposed ethical guardrails (Shepard: 06:45).
• On Autonomous Weapons:
  "Militaries around the world are really confronting this question about where, how far down humans should actually cede control to this technology, to AI..." (Mike Shepard, 07:48).

C. Broader AI Adoption and the State of the Union

• AI as National Priority: Trump’s administration pushing ‘AI-first’ for economic revitalization — chip plants, data centers, tariffs to spur domestic AI investment.
• Impending Address: Expectation for the president to address both the benefits and social anxieties around AI (Mike Shepard, 11:00):
  "It is truly a centerpiece of his economic program and plan even as it does raise some affordability issues when it comes to the demands on power."

D. Foreign Policy Flashpoints: Ukraine & Iran

• Ukraine War:
  • Four years in, U.S. stalemated on sanctions and negotiations, with casualties climbing and no end in sight.
  • Rick Davis: “There’s only one resolve to this, and that is the defeat of Vladimir Putin and the Russian war machine. … this administration owes it to the American people... to really slam home painful sanctions on the Russian economy.” (17:06)
  • Jeannie Shan Zaino: “Had Ukraine agreed to what Joe Biden quite frankly talked them out of... they would still have those four oblasts... Now, they are risking losing their access to the sea and becoming a rump state.” (18:45, 23:16)
• Iran: Warnings about being "spread thin" (Davis) and the risks of escalation in the Middle East, including the dangers of Iran developing a nuclear weapon (21:10).

---

3. Expert Deep-Dive: Gregory Allen (CSIS, Ex-DoD AI Official) [30:10–41:59]

The Pentagon/Anthropic Standoff — Technical & Policy Realities

• Anthropic’s Unique Role:
  • "Claude is the only active AI model that is really already providing meaningful military and intelligence advantage using advanced AI capabilities... In the raid on Maduro... that would not have been possible had Claude not been involved." (Gregory Allen, 30:28)
  • DoD user base “loves Claude…” and none of Anthropic's restrictions have yet been triggered, even as they've already walked back many demands.
• Pentagon’s Escalating Demands:
  • DoD wants Anthropic obligated to “make their tools available for all lawful uses.” Allen concedes that military flexibility is important but says the Pentagon’s threat is “dramatic escalation.”
  • "To take a domestic AI champion ... and light it on fire over something like this. There is a better way to resolve this dispute than the absolutist stance the administration is taking." (32:12)
• Supply Chain 'Risk' Designation — Nuclear Option:
  • This would not only eliminate Anthropic’s $200m DoD contract but could devastate its entire commercial viability by scaring off any other company that supplies the government (35:11).

Policy and Labor Market Impacts

• Agentic AI & Automation:
  • Rapid convergence where “agentic AI” (AI agents that can code and build software) could make Software-as-a-Service (SaaS) models obsolete, fundamentally shifting U.S. software industry towards more bespoke, in-house capabilities, eliminating bottlenecks (36:34–37:38).
• Risks for Labor:
  • AI increases productivity like the tractor did, but warns of “Grapes of Wrath” scenarios — occupations that can’t adapt may simply vanish:
    • "The population of America's workforce that was working on farms was something like one out of every three laborers in the 1920s. Today, it's closer to one out of every 100, even though we produce way more food. … But you know who didn't retrain to find new skills? Horses." (38:22)
  • If AI surpasses humans at most tasks (“the threshold”), “Why would people be employing people?” (39:31)
  • "If you listen to the leading lights of AI... Dario thinks we'll cross it in 18 to 24 months." (39:41)
• Advice to Workers:
  • "If you work in a field where there is not a lot of training data ... or it is not amenable to this automated testing, it's going to be a lot longer before that sort of thing can actually be automated. But over a hundred year time frame, it's very hard for me to say why anyone would be safe." (41:06)

---

4. Notable Quotes & Memorable Moments (by Timestamp)

• On Pentagon–Anthropic tension:
  “There is clearly a bit of friction and a bit of difference over where the boundaries should be set, Joe, when it comes to use of artificial intelligence technology in the military.”
  — Mike Shepard, [04:07]

• On AI for war and precedent:
  “Grox addition through Xai... adds a layer of competition for work within the Pentagon.”
  — Mike Shepard, [05:33]

• On AI policy ambiguity:
  “They do not sanction or condone any illegal use of artificial intelligence technology... What they don't want though, is they don't want additional restrictions on use of this technology from companies that go beyond whatever the letter of the law would say.”
  — Mike Shepard, [06:45]

• On Anthropic's users:
  “The user base within the Department of Defense loves Anthropic, loves Claude, and says that their restrictions on usage... have never been triggered. ... [Anthropic] have already walked back so many of their requests.”
  — Gregory Allen, [30:28]

• On the threat of blacklisting Anthropic:
  “That could make Anthropic a non starter for a whole huge segment of the American economy and could be almost fatal to their business. ... You do not want to take one of the crown jewels of your industry and light it on fire over something like this.”
  — Gregory Allen, [32:12, 35:11]

• On historical analogy to tractors & horses:
  “Humans in that story were able to go work in manufacturing sectors ... because they could retrain ... you know who didn't retrain to find new skills? Horses. ... There was no comparative advantage.”
  — Gregory Allen, [38:22]

• Labor market reality check:
  “If you work in a field where there is not a lot of training data ... or ... not amenable to this automated testing, it's going to be a lot longer before that sort of thing can actually be automated. But over a hundred year time frame, it's very hard for me to say why anyone would be safe.”
  — Gregory Allen, [41:06]

---

5. Important Timestamps for Key Segments

• 01:01–03:37: Introduction to the Pentagon–Anthropic conflict
• 04:00–11:57: Discussion with Mike Shepard on AI in defense, policy tensions, implications of new competitors
• 15:42–26:21: Ukraine war debate, geopolitical context, and State of the Union preview (Rick Davis, Jeannie Shan Zaino)
• 28:15–41:59: Gregory Allen segment — technical background, policy implications, existential anxieties around AI, labor, & economic transition

---

6. Episode Takeaways and Implications

• High-Stakes Power Play: The Pentagon's hardline with Anthropic has implications beyond a single contract, potentially chilling tech/government collaboration or dictating a new standard for AI vendor relations in security.
• AI’s Unstoppable March: The emergence of competitive LLMs for classified US government use signals the start of an arms race that is both technological and ethical.
• Societal Jitters: Even as political leaders frame AI as an economic engine, top experts openly admit uncertainty about how to protect the labor market, retrain displaced workers, and regulate truly transformative AI.
• Unresolved Ethical Boundaries: Both government and industry struggle to draw the line between maximizing operational capability and upholding democratic values.

---

7. Summary Table of Key Participants and Positions

| Speaker | Role/Expertise | Main Points/Quotes | |--------------------|--------------------------------------------------|---------------------------------------------------------------------------------------------------------| | Joe Mathieu | Host, Bloomberg | Guides the conversation around Pentagon–Anthropic, Ukraine, AI economy | | Kailey Leinz | Host, Bloomberg | Reports on breaking Axios coverage of Pentagon ultimatum | | Mike Shepard | Tech Editor, Bloomberg | Explains Pentagon/Anthropic policy standoff, AI military adoption trends | | Gregory Allen | Senior Advisor, CSIS (ex-DoD AI) | Gives insider context on Anthropics’ work, supply chain risk dangers, historical analogy to automation | | Rick Davis | GOP strategist | Urges tougher anti-Russia sanctions, skepticism on ending Ukraine war | | Jeannie Shan Zaino | Democratic analyst, Harvard | Critiques negotiation failures on Ukraine, consequences for settlement, concerns for Ukraine's future |

---

Overall Tone & Language:
The conversation is urgent, analytical, and candid, reflecting both public reporting and behind-the-scenes policy anxieties. While hosts stay measured, expert guests (especially Gregory Allen) speak plainly about risks, stakes, and industry realities.

---

For listeners/readers who missed the episode:
This summary provides a comprehensive understanding of the Pentagon’s escalating confrontation with Anthropic, the larger governmental/industrial scramble for AI dominance, the ripple effects on national security and geopolitics, and the profound societal uncertainty swirling around the rapid advance of artificial intelligence.

Transcript

[Kailey Leinz]

Being a small business owner isn't just a career, it's a calling. Chase for Business knows how much heart and effort go into building something of your own. Manage all your business finances, from banking to payments to credit cards, all in one place with Chase's digital tools. Plus access online resources designed to help your business thrive. Learn [e-mail] business chase for business Make More of what's yours the Chase Mobile

[Bloomberg Announcer]

app is available for select mobile devices.

[Kailey Leinz]

Message and data rates may apply. JP Morgan Chase Bank Naomi Member FDIC Copyright 2026 JPMorgan Chase Co.

[Bloomberg Announcer]

Bloomberg Audio Studios Podcasts Radio news. You're listening to the Bloomberg Balance of Power podcast. Catch us live weekdays at noon and 5pm Eastern on Apple CarPlay and Android Auto with the Bloomberg Business app. Listen on demand wherever you get your podcasts or watch us live on YouTube.

[Joe Matthew]

Just got one more piece and this will be kind of trickling out through the day. As Tyler mentioned, we don't have the full list of those who are going to be in the First Lady's box this evening, but Caroline Levitt, the press secretary on Twitter just now, Erica Kirk will be one of the President's special guests at the State of the Union. She writes, the president will call on Congress to firmly reject political violence against our fellow citizens with Charlie Kirk's widow in the chamber. Just posted just now. And as I mentioned, we'll be learning more, of course. Jack Hughes in town, do we know? I checked his Twitter. I don't. I don't see any pictures of marble. He's got to come here before he goes to the dentist, right? For the implants? We'll find out. Producer James gonna get to the bottom of this for us. We also have to keep tabs on what's happening on the other side of the Potomac today. This is a big story that I want to flag for you as the Bloomberg listener, because we have a pretty good sense of what's important, right? This is why we get together every day and talk. It's not just the state of the UN on the other side of the Potomac, inside the Pentagon, the Secretary of Defense is set to meet with and could right now be meeting with the big boss at Anthropic, the AI company. This has a lot to do with some of the guard rails that we have talked so much about that some believe should be on artificial intelligence. And the big boss at Anthropic, Amadei is his name, right? Dario Amadei, has spent a lot of time writing and talking about some of the moral and ethical conflicts that this new technology presents. He does not want his technology. And by the way, Anthropic is the, well, at least until today was the only major AI model that was used for in classified systems at the Pentagon. That has just changed now. Grok has that classification as well and apparently Google's working on it for Gemini. Anthropic does not want its product to be used for domestic mass surveillance or, or for human free weapons. Right. We don't want to have the Terminator with our software. The Pentagon wants them to bring the guardrails down though. And this is going to be a complicated situation if they have to uninstall effectively Anthropic from all of the systems in the Pentagon, some of which were used in the capture of Maduro in Venezuela. This is where we start our conversation with Mike Shepard, Bloomberg Senior editor for Technology and Strategic Industries. And what is a true story, a true example of Washington and Wal street meeting together here, policy and corporate America colliding. Mike, it's great to see you.

[Mike Shepard]

Thanks, Joe.

[Joe Matthew]

Does anybody seen Jensen Huang today? Is he in Washington?

[Mike Shepard]

Well, that was one question you were asking about Jack Hughes. I was immediately thinking, wait, did the invitation for the first ladies box the Nvidia CEO Jensen?

[Joe Matthew]

He's going to stand up in the gallery tonight, isn't he? Do we know where he is?

[Mike Shepard]

We don't know where he is. We've been asked.

[Joe Matthew]

I'm just feeding speculation. So you were already.

[Mike Shepard]

Of course, yeah.

[Joe Matthew]

That'd be a fun surprise. I don't know if Mr. Amadei is invited because we're told this is not a fun meeting at the Pentagon.

[Mike Shepard]

That is our sense that it will not be a friendly atmosphere. The company has said that it is engaged in productive and good faith conversations with defense officials. And yet the tone from the Defense Department has been a little bit different. What they have characterized it as publicly in a statement last week from their spokesman Sean Parnell, is that the relationship between, not just the contract, the relationship between Anthropic and the Defense Department was under review. And they've made clear that they don't want companies to go in there not being willing to support all war fighting efforts. And it's unclear exactly where this rupture and breakdown took place and exactly how these communications and concerns have been conveyed from one to the other. We're getting snippets of it publicly and in reporting with, you know, on both the company side and on the Defense Department side. But there is clearly a bit of friction and a bit of Difference over where the boundaries should be set. Joe, when it comes to use of artificial intelligence technology in the military, I

[Joe Matthew]

thought this had the potential to be precedent setting. Maybe you will tell me that it still does. But when we see Grok being cleared to to for use in DoD classified systems, is this as important as it was a day ago?

[Mike Shepard]

Well, it is in a way, because Grox addition through Xai it adds a layer of competition for work within the Pentagon and within the broader national security apparatus of the US Government, which includes all the intelligence agencies as well. And so you have to think is that you have more of these companies breaking into this space, gaining the qualifications and clearances to do so. Of course, that's competition for Anthropic, but it's a lot of other business for this sector of Washington that, you know, often operates in the shadows. I mean, a lot of the budgeting even is kind of off the books and, you know, regarded as, you know, classified or confidential. And it is nonetheless consequential in terms of its execution and what the technologies and the products are being used for. And that is one of the key questions that Anthropic and others really have been raising as well.

[Joe Matthew]

So I know the Pentagon is not new to AI They've been working on this technology for years. But are they essentially building the plane in flight when it comes to its own policy? Do we know where the line is for the Pentagon when it comes to, for instance, mass surveillance or human free weapons?

[Mike Shepard]

Well, that is a great question, Joe, about where the Pentagon stands in terms of its policy. And last month, and this may really be what sort of accelerated and caused to erupt this break between the company and the Defense Department. They published an AI strategy and they really want to hit the gas on adoption of artificial intelligence and really make the US Military an AI first force, as they put it. One thing they make clear in the document, though, is that they do not sanction or condone any illegal use of artificial intelligence technology. They everything would have to comply by, you know, by U.S. law, by the laws of war and everything else. And what they don't want, though, is they don't want additional restrictions on use of this technology from companies that go beyond whatever the letter of the law would say.

[Joe Matthew]

The Geneva Convention obviously didn't get to this. And the idea of autonomous weaponry brings in a whole new challenge, doesn't it?

[Mike Shepard]

Well, it does. And back to the question of mass surveillance of US citizens domestically, the Pentagon would say, look, we don't even do that. That's just not who we are. And likewise with autonomous weaponry, you know, they haven't really gotten to the point where, look, we will do this. And militaries around the world are really confronting this question about where how far down humans should actually cede control to this technology, to AI and, you know, the last 600 meters in any conflict where, you know, you really have that, that moment, you know, whether they must decide to take out an individual, enemy, soldier, combatant, or groups of combatants, that has always been the purview of, you know, people. And whether that will change and how that will change is really still an evolving question around the world, not just

[Joe Matthew]

here in the US the last 600 meters, that's going to be the name of your next bookshop.

[Mike Shepard]

I think it may be the name of somebody's book all day, honestly.

[Joe Matthew]

Seth Moulton, gosh, it was two years ago, he wrote an op ed calling for an AI Geneva Convention, that it wasn't the US he was worried about, it was everybody else. That nothing would keep Russia, for instance, not that it would sign off on a Geneva Convention or other nations from putting armed robots in the field or autonomous weaponry and so forth. This is a Pentagon that's not waiting around to find out, though, to your point, right, Pete Hegseth doesn't want to be on, on the back foot when it comes to this technology. So when you start including Google, Gemini Elon Musk's X AI, where are they on possible governors? Do we have any sense of where they draw the line?

[Mike Shepard]

And that's a. That's another good question. And they've also had their own. XI is maybe a little bit different in the sense that it is Elon Musk's company. And we have seen that, you know, they take a different approach when it comes to guardrails. Overall, we've certainly seen that with X. And it's unclear how Grok. We all remember Mecca Hitler from last year, how it would be controlled by the, you know, by internally, by its own policies. Now, Google, of course, has done much more government work over the years, not in AI necessarily. And they have tried to set certain guardrails. And we saw years ago even protests by Google employees about the company's work with the defense and intelligence establishment come

[Joe Matthew]

a long way from there.

[Mike Shepard]

We have come a long way from that there, Joe. And this does give rise to a whole new set of questions about where the companies and their rank and file and of course, their leadership, who have really approximated themselves so much to the Trump administration since, you know, the president's return to power. It'll be interesting to see how those boundaries get set in this new environment.

[Joe Matthew]

This is some really high minded stuff and I want to let everyone know that in our next hour, actually we're going to explore these ideas more with Gregory Allen from csis, who's a great mind and in fact worked on the Pentagon's AI program. Bring it home for me for tonight. Shep, does the president get his arms around AI in the State of the Union address and does he have new proposals to share?

[Mike Shepard]

Well, Joe, he has had his arms around AI since taking office. You remember in his first full day in office he was there unveiling this Stargate initiative with Sam Altman, with, with other big investors and every step of the way and he has framed his tariff program, the cornerstone is of economic agenda as a way to bring a lot of that tech heavy AI, heavy investment here into the U.S. it's not just the data centers, it's also the chip making plants that are producing the processors needed to power AI. It is truly a centerpiece of his economic program and plan even as it does raise some affordability issues when it comes to the demands on power. So he will definitely address this in some fashion. Talk about tariffs have spurred investment and talk about the tech side of, of what that brought.

[Joe Matthew]

Not lost on me that tomorrow is in video earnings right after the bell. So if Jensen Huang's in the gallery, he's got to get back to the west coast, presumably for the conference call, right?

[Mike Shepard]

It's so. Well, he could do it almost from anywhere. But yes. And, and it is a red letter week on the calendar, which Jensen is in the gallery.

[Joe Matthew]

Is that worth five points on the stock the next day? That's going to be my question. I think you're not allowed to answer that.

[Mike Shepard]

I'm not allowed to answer.

[Joe Matthew]

Shep is with us on State of the Union Day in Washington. I know you've survived a couple of them. It's great to have you here to be Shepard live from Bloomberg in Washington, our senior editor for technology and strategic industries. We've got a lot more coming, including our political panel there with us for the long haul, Jeannie Shan Zaino and Rick Davis. Bloomberg Politics contributors are on the way a half hour from now. Kailey Leinz will jump in the pool as well. Stay with us on balance of Power. We'll have much more coming up after, after this.

[Public Ad Narrator]

Support for the show comes from public. Lately it feels like there are two types of investing platforms. Some are traditional brokerages that haven't changed much in decades and others feel less like investing and more like a game. Public is positioned differently. It's an investing platform for people who are serious about building their wealth on public. You can build a portfolio of stocks, options, bonds, crypto without all the bugs or the confetti. Retirement accounts? Yep. High yield cash? Yes again. They even have direct indexing. Public has modern design, powerful tools and customer support that actually helps go to public.com market and earn an uncapped 1% bonus when you transfer your portfolio. That's public.com market and paid for by

[Mike Shepard]

Public Holdings Brokerage services by Public Investing member FINRA SIPC Advisory services by Public Advisors SEC Registered Advisor Crypto Services by ZeroHash All In Investing involves risk of loss. See complete [e-mail] disclosures

[Bloomberg Announcer]

you're listening to the Bloomberg Balance of Power podcast. Catch us live weekdays at noon and 5pm Eastern on Apple CarPlay and Android Auto with the Bloomberg Business app Listen on demand wherever you get your podcasts or watch us live on YouTube.

[Joe Matthew]

It's a big one today. The super bowl of politics. State of the Union here in Washington, D.C. the president will speak around 9pm Eastern as he enters the chamber. We'll have special coverage underway long before then with our signature panel. Bloomberg Politics contributors Jeannie Shan Zaino and Rick Davis will be with us and glad to say they're with us throughout the day as we gear up for the big speech. The president scheduled at least to be having lunch with the network news anchors right now at the White House, and he's got a couple of things to do. They'll be polishing the speech, then they need to load it into the prompter. And of course, guests will be arriving to spend time with the President, the First Lady. Everybody gets to the other end of Pennsylvania avenue in the 8 o' clock hour. You'll hear and see the motorcade on its way up there. As part of our special coverage. When the president arrives, the carriage entrance will follow him straight into the chamber for what's going to be maybe a record. You're ready for this. President@ 99 minutes last year set a record and he is already saying it's going to be a long one because he's got a lot to talk about tonight. So yeah, we could be in uncharted territory as we mark four years of war. Here's one that not a lot of people are talking about today because there's so much else to get to, whether it's starting a new war in Iran or dealing with the tariff picture the president's efforts to address affordability. Four years of war in Ukraine. As of today, the war the president promised to end on his first day back in office with nearly 2 million people killed or injured in the biggest land war in Europe since World War II. The president did speak about it a year ago in his address to a joint session of Congress.

[Public Ad Narrator]

Listen.

[Joe Matthew]

Regarding the agreement on minerals and security, Ukraine is ready to sign it at any time that is convenient for you. Simultaneously, we've had serious discussions with Russia and have received strong signals that they are ready for peace. Wouldn't that be beautiful? It's time to stop this madness. It's time to halt the killing. No, that, of course, is not how it played out. And talks have continued with no evidence that Vladimir Putin or those negotiating for Russia have any interest in acquiescing to any of the conditions that would satisfy Ukraine. So let's bring him in. Our panel is in for the long haul. Rick Davis is Republican strategist and partner at Stone Court Capital. And Jeannie Shan Zaino is our Democratic analyst and democracy visiting fellow at Harvard Kennedy School's Ash Center. Happy State of the Union Day to both of you. Rick, I'll start with you on Ukraine. The president, according to Bloomberg News, is now pushing for a deal before he hosts the 250th birthday celebrations for American independence. He wants a deal by the 4th of July. And I don't know why he's bypassing Easter. We all, we remember the big reopening for Covid that he wanted to coincide with the Easter holiday. Is he capable of ending the war in his second year back in office?

[Rick Davis]

I really doubt that he can do that. Obviously, we all want this war to end. Four years is too long for this conflict. But I think the primary thing that's been missing in that period of time, both with the Biden administration and the Trump administration, is a real clear eyed understanding that Vladimir Putin has absolutely no interest in settling this war. And I think there's lots of evidence to indicate that there's only one resolve to this, and that is the defeat of Vladimir Putin and the Russian war machine. And that will come when real sanctions are applied, that there is a real effort to try and squash the Russian economy. You know, there's as much oil being pumped and sold out of Russia today than when they started the war four years ago. What is wrong with that picture? Why can't we get our allies and, and friends to really impose strict and painful sanctions? There's a bill in the Senate sponsored by Lindsey Graham and others, 85 senators have signed on to this bill to really slam home painful sanctions on the Russian economy to try and end this war. This administration owes it to the American people who have invested in this. And the Russian people are fighting for their freedom to pass that bill and sign it into law.

[Joe Matthew]

Really interesting, Jeannie, we heard from the UK this morning targeting the Russian oil fleet with biggest sanctions in years. Is the President of the United States ready to do something like that, to make an announcement like that tonight?

[Jeannie Shan Zaino]

I don't see any evidence of that. In fact, we've been moving in exactly the opposite direction. And of course, we are four years in, by some counts, 2 million or more casualties. And there doesn't look to be any end in sight to this conflict. And the reason is pretty simple because neither side, and I think we should underscore neither side, it's not just Russia, it is Ukraine as well, are willing to come to the negotiating table and to give up their demands. And of course, had Ukraine agreed to what Joe Biden quite frankly talked them out of with Boris Johnson in 2022 and that 126 page deal that was on the table, they would still have those four oblasts. They've lost land since then. And the probably really difficult reality is they would be in a good position now. Not it's a terrible position, but it's the best they're probably going to get to make a deal now to give up some of that land in return. But we've heard for Zelensky, he's unwilling to do that. So I don't see the President of the United States stepping out at this point. And of course, without the US Europe is not in a position to push back. And so we are at a stalemate. This thing is going to be won or lost on the battlefield. And at this point, it looks like Russia has the upper hand and they know it. They're talking when you listen to Russian media about a new Russia, that new Russia includes a whole host of land in Ukraine. And that is something that Ukrainians are going to have to make some really difficult decisions about. If you just look at it from one perspective, they don't have the manpower in the military to combat Russia for a prolonged war at this point.

[Joe Matthew]

Well, I do wonder what the president will say about this beyond calling parties to the table. Rick, you never know exactly which President Trump you're going to get on this issue. He most recently said that it was up to Volodymyr Zelensky to get back to the negotiating table. That it was Ukraine that was somehow dragging its feet on a deal. But he's also got the matter of Iran to deal with. And I'm wondering how you think he balances these two while calling for peace in one war and threatening the possibility of starting another.

[Rick Davis]

Yeah. You just got to say for the record, there's only one person that has been not negotiating any terms other than complete capitulation of the rush of the Ukrainian people, and that's Vladimir Putin. So if there's anybody to blame, there's only one person to blame. And the fact that the administration can't seem to understand that and waste their time trying to negotiate with somebody like him is unfathomable. You're right. I mean, we start to look like we're getting spread thin with another conflict in the Middle East. Iran, very bad group of people. They deserve to be toppled. There needs to be support in the country to do so. Colleges are back now. Students are already starting to protest again. The president's going to have to make some tough decisions. This is not Venezuela. There's real resistance. It could be destabilizing to the Middle East. And I think this is probably one of the most important, if not difficult, decisions that the president's going to make in order to try and keep the Middle east from having a nuclear Iran. Look, Iran wants to be North Korea of the Middle East. We see how that's worked out for people who have been under the thumb of the North Koreans, the South Koreans, the Japanese, other free people in the region who have to forgo all these nuclear tests. And it's just. It's outrageous. Could you imagine that happening, you know, with Iran in the Middle East? So it's, it's, it is an important decision by the president, but one that we think he's not taking lightly. But nobody has a clue how this is going to resolve itself.

[Joe Matthew]

Well, that's saying a lot. Jeannie, how does Abigail Spanberger manage all of this? She's going to talk about her national security bona fides. Right. She was a CIA officer and of course spent time as a member of Congress dealing with these issues. She's supposed to be talking about affordability, right? This is. The moderate Democrat is going to talk about domestic issues. Does she spend time on this?

[Jeannie Shan Zaino]

I think most of her time is going to be spent contrasting her vision, this moderate democratic vision, with that of Donald Trump, beginning with chaos and costs, to your point, about affordability. But given her background in national security, in foreign policy, we may hear her talk about this. I think if she does, it'll be in the context of issues of chaos, issues of promises unkept by this president who said he wasn't going to get us involved in foreign entanglements. And I think the president owes it to us on the issue of Iran to explain his thinking. Why are we, why have we built up as much as we have military over there? And what is the end goal? What does he hope to achieve? And just to get back to Ukraine and Russia for a moment. As much as we may like to pretend there's only one side here, there are two sides here and it is Russia versus Ukraine. And Vladimir Zelensky is in a horrible position, as is Ukraine. They were invaded in 2014, again in 2022. But the reality on the battlefield is had they agreed in 2022 to that 126 page document, those four oblasts that Russia has taken would not have been taken and they would be in a better position at this point. They are risking losing their access to the sea and becoming a rump state. There is no thinking person who wants them to end up like that. And that's where hard decisions have to be made. And so you can cast blame on anybody you want, but the reality is both sides are going to have to concede or this thing is going to end on the battlefield. And by all estimates, if the US doesn't back Ukraine on this in a much more concerted way than the President has shown he wants to, it is going to be unlikely that they can prevail on the battlefield because Europe's not in a position to do it and they have been unwilling to do it. So it is a very tough situation for Ukraine.

[Joe Matthew]

Rick, the president's going to have to play triage with the topics unless he wants to talk until tomorrow here. And knowing that Abigail Spanberger is going to be heading to Leesburg the very next morning to kick off a Democratic retreat, they're going down to Lansdowne Resort. Rick, to talk about fighting for an affordable America. That's the theme of the event. How does Donald Trump proportion the speech between foreign policy and domestic agenda in our remaining moments here?

[Rick Davis]

Well, I think, you know, the president has no limitations. He may be talking until 1 o' clock in the morning, you know, and so I think he's going to try to include everything. I mean, he's a president who ran on America first and yet at the same time has spent an enormous amount of time on foreign policy, national security issues, trade, things that are not considered core America first issues. And I'M sure there are advisers telling him we've got to get back to that theme, so we'll see how he portions it out. But my guess is we get a

[Joe Matthew]

little bit of everything. No, we're going to be running the office pool on length here, guys, so think about that before. Our second hour was 99 minutes last year. Everybody in the control room is hoping he's good till at least one o' clock in the morning, but we'll figure that out together in real time tonight. Stay with us on Balance of Power. We'll have much more coming up after this.

[Public Ad Narrator]

Support for the show comes from Public. Public is an investing platform that offers access to stocks, options, bonds and crypto, and they've also integrated AI with tools that can assist investors in building customized portfolios. One of these tools is called Generated Assets. It allows you to turn your ideas into investable indexes. So let's say you're interested in something specific, like biotech companies with high R and D spend, small cap stocks with improving operating margins, or the S&P 500 minus high debt companies. Chances are there isn't an ETF that fits your exact criteria. But on Public you just type in a prompt and their AI screens thousands of stocks and builds a one of a kind index. You can even backtest it against the S&P 500. Then you can invest in a few clicks, go to public.com market and earn an uncapped 1% bonus when you transfer your portfolio. That's public.com market ad paid for by

[Mike Shepard]

Public Holdings Brokerage Services by Public Investing member finra SIPC Advisory Services by Public Advisors SEC Registered Advisor crypto services by ZeroHash. Sample prompts are for illustrative purposes only, not investment advice. All investing involves risk of loss. See complete [e-mail] disclosures.

[Bloomberg Announcer]

You're listening to the Bloomberg Balance of Power podcast. Catch us live weekdays at noon and 5pm Eastern on Apple Car Play and Android Auto with the Bloomberg Business app. You can also listen live on Amazon Alexa from our flagship New York station. Just say Alexa play Bloomberg 11:30.

[Kailey Leinz]

I'm Kailey Leinz alongside Joe Matthew here in Washington, where if it were any other day other than being State of the Union Day, we probably would have dedicated a lot more time already this hour to a meeting that took place at the Pentagon today between the Defense Secretary Pete Hegseth and the CEO of Anthropic. What's in question here is Claude and its cooperation, or lack thereof, with the DoD when it comes to concerns that at least Anthropic has around safeguards around its AI modeling. Apparently the Defense Secretary. This is according to reporting from Axios, gave Anthropic until Friday evening to give the military unfettered access to its AI model or face harsh penalties. The harsh penalties we know, Joe, that could be in question here are including declaring Anthropic a supply chain risk or even, according to Axios, invoking the Defense Production act to force the company to tailor its models to the military's needs.

[Joe Matthew]

The CEO at Anthropic, Dario Amadei, has spoken eloquently and in some cases in a very scary fashion about the risks and the challenges that AI brings to the table here. And this is for him a part of a moral question about using the technology for mass surveillance or for human free weapons, autonomous weaponry. And this is the conversation apparently very tense meeting with Hegseth. He's got till Friday now knowing, and this is important, as the New York Times scoop today. Grok has been cleared. The XAI tool to be used in classified systems. So there could be a replacement if Anthropic is shown the door here. And it's where we start our conversation with one of the most important minds in the business on this technology. Gregory Allen is back, senior advisor at the Wadwani AI center at csis, the center for Strategic and International Studies. I will remind you he was Director of Strategy and Policy at the Department of Defense Joint Artificial Intelligence center and would have been in the room for this meeting if he still had that job today. It's great to have you back with us on Bloomberg TV and radio.

[Gregory Allen]

Great to be here.

[Joe Matthew]

I don't want to turn you into someone who needs to make moral judgment on the air, but talk to us about how important important it is to have this access and what exactly the Pentagon is doing with this software. We're not at the point of having human free weapons. What is Pete Hegseth asking Anthropic to do?

[Gregory Allen]

Well, I think there's a few things worth unpacking here. The first of which is that unclassified DoD military computer networks. Claude is the only active AI model that is really already providing meaningful military and intelligence advantage using advanced AI capabilities. For example, we've seen in reporting by the Wall Street Journal that the raid on Maduro was enabled by offensive cyber capabilities. That would not have been possible had Claude not been involved, had Anthropic not been involved. And since anthropic won that $200 million contract to work with the Department of Defense, they have been leaning forward among all the frontier AI companies in most aggressively support supporting national security. So this dispute comes at an awkward time because on the one hand, the user base within the Department of Defense loves Anthropic, loves Claude, and says that their restrictions on usage, at least from the conversations that I have been having, have never been triggered. There has never been a time where Anthropic said no. And it's worth pointing out that when Anthropic first negotiated this contract back in July 2025, the DoD was comfortable with a much longer set of restrictions than what Anthropic is asking for right now. They have already walked back so many of their requests. Now all they're saying is we don't want our tool being used for mass domestic surveillance, and we don't want our tools being used for autonomous lethal weapons. Without a human on the loop, that's a pretty modest set of requests. And the DoD views that is so unreasonable that they're now threatening to designate Anthropic a supply chain risk, which is a dramatic escalation.

[Kailey Leinz]

So do you think it's unreasonable? Would this severely hamper the ability of the DoD to keep the country safe?

[Gregory Allen]

So, to be fair to the dod, what they're saying is they want a contract that allows Anthropic, or, sorry, requires Anthropic to make their tools available for all lawful uses. Now, why might they want that? You can understand that the DoD, when lives are on the line, when national security is on the line, they do not want to be having a debate with their company providers about, can we use it for this, can we use it for that? They need flexibility. And I understand that. However, the DoD already agreed to a much more restrictive set of terms back in July 2025. And so while I. I'm sympathetic to the DOD in saying it needs flexibility in these contract terms, what I think is an unreasonable escalation is threatening to designate Anthropic as a supply chain risk. That's what we do when we find out companies are secretly have Chinese or Russian ownership to take a domestic AI champion at a time when the White House is saying that the AI race with China is equivalent to the space race during the Cold War with the Soviet Union. You do not want to take one of the crown jewels of your industry and light it on fire, fire over something like this. There is a better way to resolve this dispute than the absolutist stance the administration is taking.

[Joe Matthew]

Well, this has been an incredible turn from 2018, when Google saw itself out here. Now these companies are breaking the door down to get in. Or is Anthropic just the the exception because Gemini is ready xi or they're all ready to replace Anthropic at a moment's notice? It seems.

[Gregory Allen]

Yes, at least on unclassified military networks. All the leading US model providers are already active on platforms like Genai Mil. But where Anthropic really stands out is these enterprise coding capabilities and through a partnership with Palantir, is made available to some of the most important use cases across the intelligence and war fighting communities. So Anthropic is right now delivering the best bang for buck. They're involved in the most high profile operations and so far the user can community really likes what they're getting and has said that since July 2025 they are not encountering these restrictions in a realistic scenario. Now, one thing I want to say about the use cases that Anthropic is currently negotiating around, for example lethal autonomous weapons. It is true that that is a capability that Russia is operating right now in Ukraine with Systems like the VT V2U unmanned drone that does operate autonomously, does use AI for lethal capabilities. So I can see why the DoD would say to a company, we don't want you to decide whether or not we can match the capabilities of our adversaries. That's a decision that we need to make. But as you're having that debate, do you really need to threaten to designate the company as a foreign supply chain risk or at least the equivalent of a foreign supply chain risk? There's a much better way to have this conversation.

[Kailey Leinz]

Well, and just remind us, Gregory, what would happen if indeed Anthropic is given that designation? It goes far beyond just eliminating the contract with the dod.

[Gregory Allen]

Yes, the biggest risk for Anthropic, which has a very large and rapidly growing base of commercial customers, billions of dollars of revenue every year, increasing exponentially every year. The big threat is not this $200 million DoD contract. The biggest threat of that supply chain risk designation is that it could prevent other companies that supply to the DoD from doing business with Anthropic. In a worst case scenario. That could make Anthropic a non starter for a whole huge segment of the American economy and could be almost fatal to their business. And that's why I think you should not even be casually throwing around threats like that. If you're trying to get startups excited about working with the DoD, do you really want to tell them that if you even dip your toe in the water, we're going to go grab you and pull you all the way in. That's not a great message to be.

[Joe Matthew]

Isn't that something? Let's, let's zoom out for a moment here. You mentioned Palantir and it has been wild just the last couple of weeks to watch the reaction in the markets from our vantage point here on Bloomberg. But even the conversation within the industry when it comes to agentic AI, we have written off entire industries since you last joined us, beginning with software. Now we're moving on to cyber. Now we're seeing Palantir fall out of bed because apparently this incredibly unique service that it once provided the Pentagon could be dreamt up by an AI bot overnight. Are we getting ahead of ourselves?

[Gregory Allen]

Well, you know, you're asking me to draw out my old MBA experiences, not my public policy experiences, but I'm happy to go there. Let me just say that the software industry of the United States and China looks very different. And it's because the United States software industry has had a shortage of software software engineers, which is why software as a service, as a business model has been really attractive. In China. They have an abundance of software engineers, so there's not really a software as a service business model. In fact, most companies just have in house developers who develop custom stuff just for them. So SaaS like Salesforce is practically non existent in China, even though companies still perform those functions. Well, what if agentic AI makes the US Software ecosystem look like the Chinese software ecosystem with an abundance of software engineers? So do you really need to pay Salesforce to use a tiny sliver of their service that isn't even customized for you when you can just bring in some agentic AI agents who can create the exact, perfect, highly customized set of software that everybody in your company wants to use. It could be a whole reorientation of the industry because that bottleneck might be broken.

[Joe Matthew]

So it sounds like you think that skepticism is warranted in this case.

[Gregory Allen]

At a minimum, every company needs to have a justification for how that tsunami is not going to drown them.

[Kailey Leinz]

Well, and so it obviously raises massive questions for the labor market. We've had Fed governors left and right talking about this in recent weeks that they don't even know how to properly assess what it's going to do to productivity, what it's going to do to the labor force. Lisa Cook suggested today she doesn't even know if monetary policy is a tool that is going to be able to offset what AI could do to the economy. Which raises putting your policy hat back on what policy tools can do that. I would assume it has to come in the fiscal realm and from the government. So to do so, does President Trump need to get on this, like tonight in the State of the Union to get ahead of this? Is it even possible still to be ahead of it?

[Gregory Allen]

I expect President Trump to have a very optimistic message about AI because this is a massive opportunity for productivity enhancement. This is equivalent to the invention of. Of the tractor. But it's worth thinking about. You know, if you've ever read the book the Grapes of Wrath, which is about farmers in that era experiencing mechanization, it was a really difficult time for them because that increase in productivity over the long term, that led to more economic growth, more jobs, but in different sectors. You know, the population of America's workforce that was working on farms was something like one out of every three laborers in the 1920s. Today, it's closer to one out of every 100, even though we produce way more food. Now, here's the thing. Humans in that story were able to go work in manufacturing sectors or in service sectors because they could retrain and go have skills that tractors didn't. But you know who didn't retrain to find new skills? Horses. And the population of workhorses in the American economy went from tens of millions to effectively zero because they could not retrain. There was no comparative advantage.

[Kailey Leinz]

Yeah, I was just about to say

[Gregory Allen]

this is the key question. Is AI a normal technology that is mostly about increasing the productivity of humans, or do we cross that threshold where AI can do everything that a human can do better and cheaper? Why would people be employing. Employing people? And if you listen to the leading lights of AI, including Dario Amade, but including his compatriots at every other company, I think. I think the skeptic of the group at this Stage is the CEO of Google, DeepMind, who is predicting will cross that threshold in 10 years. Dario thinks we'll cross it in 18 to 24 months.

[Joe Matthew]

Do you go to bed at night thinking that you are working on the very technology then that will spell our demise?

[Gregory Allen]

You know, there's a line that Jeff Bezos said in the rise of the Amazon as a book retailer, the earlier incarnation of Amazon. And what he said is that Amazon is not putting brick and mortar book companies out of business. The future is putting brick and mortar book companies out of business.

[Joe Matthew]

You're not making.

[Gregory Allen]

Amazon is merely the herald of the future. And I think that there is, you know, a real chance that AI actually meets all the hype. And we actually are living in that transformative moment. And it's crazy. Because every generation says that they're living through some radical transformation, transformative moment. But this one might really be different.

[Kailey Leinz]

Okay, so for those who are driving in their car right now, listening or watching us on TV and thinking, oh my God, oh my God, my job is not going to be safe, can you provide them any real reassurance? If you're someone who works on spreadsheets all day or looks at legal paperwork, I mean, the kind of of jobs that we're told artificial intelligence easily could replace.

[Gregory Allen]

So I would say you can sort of categorize jobs based on how difficult or easy it is to automate. It's not an accident that Claude code is kind of the breakout moment for agentic AI. And that's true because the Internet is full of extremely high quality training data, like everything on GitHub for here's how to do incredibly high quality coding work. But it's also that code is amenable to instant verification, which means that Claude can generate an answer and then run a test. Does this program actually do what I'm trying to get it to do? And that means it can generate its own training data by running these tens of millions of tests. But if you work in a field where there is not a lot of training data available on the Internet or it is not amenable to this automated testing, it's going to be a lot longer before that sort of thing can actually be automated. But over a hundred year time frame, it's very hard for me to say why anyone would be safe.

[Kailey Leinz]

All right, Greg Allen, scaring us all from this year, international studies. Thanks for being here on Balance of Power on Bloomberg TV and Radio.

[Joe Matthew]

Thanks for listening to the Balance of Power podcast. Make sure to subscribe if you haven't already at Apple, Spotify or wherever you get your podcasts. And you can find us live every weekday from Washington D.C. at Noontime [e-mail]

[Gregory Allen]

this is Matt Rogers from Los Culturistas with Matt Rogers and Bowen Yang. This is Bowen Yang from Las Culturistas with Matt Rogers and Bowen Yang. Hey, so what if you could boost the WI fi to one of your devices when you need it most? Because Xfinity WI fi can. And what if your WI fi could fix itself before there's even really a problem? Xfinity is so reliable. It does that too. What if your WiFi had parental instincts? Xfinity Wi Fi is part nanny, part ninja, protecting your kids while they're online. And finally, what if your wifi was like the smartest WI fi? Yeah, It's WI Fi that is so smart it makes everything work better together. Bottom line, Xfinity is smart and reliable. You deserve the peace of mind of having WI Fi that's got your back. Xfinity.

[Mike Shepard]

Imagine that.

[Gregory Allen]

Well, the holidays have come and gone once again. But if you've forgotten to get that special someone in your life a gift, well, Mint Mobile is extending their holiday offer of half off unlimited wireless. So here's the idea. You get it now.

[Joe Matthew]

You call it an early present for next year.

[Gregory Allen]

What do you have to lose? Give it a [e-mail] Switch limited time

[Kailey Leinz]

50% off regular price for new customers. Upfront payment required $45 for three months, $90 for six month or $180 for 12 month plan taxes and fees. Extra speeds may slow after 50 gigabytes per month when network is busy. See terms.