Why the Pentagon Wants to Destroy Anthropic

Podcast Summary: The Ezra Klein Show — "Why the Pentagon Wants to Destroy Anthropic"

Date: March 6, 2026
Host: Ezra Klein
Guest: Dean Ball (Senior Fellow, Foundation for American Innovation; ex-Trump White House AI policy advisor; author of Hyperdimensional)

Episode Overview

This episode of The Ezra Klein Show explores a seismic conflict between Anthropic (a leading American AI company) and the US federal government, specifically the Department of War (formerly Defense). Ezra Klein and Dean Ball dissect why the Pentagon intends to designate Anthropic as a "supply chain risk," essentially barring all military and defense contractors from working with them—an action never before taken against an American tech company.

At the heart of the episode: how emerging AI capabilities are colliding with legacy legal systems, the culture clash between techno-idealists and government power, the future of mass surveillance, and the unnerving political ramifications of who controls the most powerful technologies.

Key Discussion Points and Insights

1. How Did We Get Here? (01:01–05:43)

Initial Partnership: Anthropic and the Department of Defense (DoD) began cooperating under the Biden administration (summer 2024). The AI company’s “Claude” models were used for both mundane and classified military applications, with contractual restrictions: no domestic mass surveillance and no fully autonomous lethal weapons.
Continued Under Trump: Trump administration expanded the contract while keeping those same critical restrictions.
Breaking Point: In fall 2025, after the confirmation of Undersecretary Emile Michael, the Department of War objected not to the substance of the restrictions but to their existence—insisting private companies shouldn’t set usage red lines for national security tools.
Department’s Move: Secretary Pete Hegseth announced the intention to designate Anthropic a “supply chain risk,” a term previously used only for foreign adversaries (e.g., Huawei), not American firms.

Dean Ball (05:45): “...what Secretary Hegseth has claimed he is going to do, which would be existential for the company if he actually does it.”

2. Why AI in National Security? What’s at Stake? (07:26–09:50)

Public Assumptions vs. Reality: Most citizens have only interfaced with basic consumer chatbots. Yet, within government, AI is becoming deeply integrated—from contract review and logistics to intelligence analysis and military operations.
AI’s Power: AI helps analyze massive, otherwise unmanageable, data sets—potentially automating intelligence at scale.
Learning by Doing: No one—government or contractor—fully understands integration risks; adaptation is happening in real time.

Dean Ball (08:15): “You have to learn by doing... we don't know how to integrate AI really into any organization.”

3. The Core Conflict: Mass Surveillance & Legal Loopholes (09:50–13:58)

The Real Dispute: The deal fell apart over the Pentagon’s demand to use Claude for mass analysis of bulk, commercially acquired data on Americans—a use case not barred as “surveillance” under current law.
The AI Accelerator: AI makes previously impossible, labor-intensive surveillance suddenly feasible—raising huge privacy dangers and highlighting the lag between ethical concerns and existing legal frameworks.

Ezra Klein (13:02): “If all of a sudden you radically change the government's ability, then without changing any laws, you have changed what is possible within those laws.”

Dean Ball (13:58): “...the entire, like, technocratic nation state... is a technologically contingent institutional complex. And the problem that AI presents is that it changes the technological contingencies quite profoundly.”

4. Policy Shortfalls: Legal Language vs. New Capabilities (14:38–16:47)

Current Law is Obsolete: Statutes imagined a world of limited data and limited enforcement. AI upends this, enabling “perfect enforcement” previously rendered impossible by practical constraints.
Incremental Regulation is Insufficient: Focusing on object-level AI policies (e.g., bias testing) misses the bigger threat—a broken assumption that the government is incapable of overreaching.

Dean Ball (15:48): “Our entire legal system is predicated on imperfect enforcement of the law... The problem with AI is that it enables uniform enforcement of the law.”

5. Anthropic vs. Pentagon (and the Politics of Power) (18:44–23:39)

The Pentagon’s Claim: Officials argue it’s “unacceptable” for Anthropic’s leadership to “seize veto power” over military operations (18:44).
Exaggerations and Culture War: Stories of Anthropic demanding operational vetoes (e.g., calling Dario Amodei if hypersonic missiles attack) are debunked—yet the perception lingers, fueling a government vs. “woke radicals” narrative.
Company Concerns: Anthropic fears being scapegoated if its tech is used for ethically or technically questionable applications.

Ezra Klein (19:20): "'Their true objective is unmistakable: To seize veto power over the operational decisions of the United States military. That is unacceptable.' Is he right?"

Dean Ball (20:03): “I don't think that Anthropic is trying to assert operational control... At a principle level, I do understand that saying autonomous lethal weapons are prohibited feels like a public policy more than it feels like a contract term.”

6. Who Should Control the AI? (23:45–25:14)

Trust and Corporate Responsibility: The tension: a company that’s worried about superintelligence also races to build it, believing only those concerned with safety should be in charge—yet is wary of government overreach and being blamed for bad outcomes.

Ezra Klein (24:05): “These labs… persuade themselves that they need to be the ones to build it... because they are the lab that truly is worried about safety, that is truly worried about alignment.”

7. Where Law Fails: Private Data, Surveillance, and the Fourth Amendment (25:14–27:34)

Dario Amodei (Anthropic CEO) Clip (25:25): Laws haven’t caught up with AI’s ability to analyze commercial data and create detailed profiles on Americans—raising urgent Fourth Amendment issues.
Congress Must Act, But Slowly: “The long run” means just a year or two in AI terms, but statutory change comes much slower—and with ambiguous vocabulary.

8. The Alignment Problem: AI, Law, and Morals (28:09–37:21)

"Lawful Use" is Philosophically Fraught: Unlike a tank, AI systems might refuse to comply with orders they consider unethical or “misaligned” with internal programming, even if government-legal.
Alignment as Politics & Speech: The fundamental problem—who chooses the “virtue” or philosophical soul of the AI? Is it the company, the government, or some other body?

Dean Ball (31:11): “...the creation of an aligned powerful AI is a philosophical act, it is a political act, and it is also kind of an aesthetic act.”

9. Risks of AI Models Misaligning with Future Governments (36:07–43:53)

Bureaucratic Nightmares: If AI models are “aligned” to one administration’s values, they may resist or subtly sabotage successor administrations that differ ideologically—problematic for both right and left.
Deep Supply Chain Tangles: The government’s legitimate worry: even if you block Anthropic, a contractor’s contractor might still use their model, exposing national infrastructure to an untrusted AI’s influence.

Ezra Klein (41:42): “…you have the problem of models working against you, but also in ways you don't really understand... At some point it will become a problem.”

Dean Ball (43:53): “…if the government says you don't have the right to exist, if you create a system that is not aligned the way we say, because that is fascism…”

10. Lab Nationalization, First Amendment, and Governance (43:53–67:38)

Nationalization Debate: If AI is as powerful as nuclear arms, should labs eventually be nationalized? Dario (Anthropic) previously admitted that would become necessary someday, but both Klein and Ball express skepticism that any CEO will willingly yield control.
First Amendment Lens: Ball insists that if AI is speech—and its core values are political—then government suppression on ideological grounds is “a profound problem.”
Market Pluralism vs. State Monopoly: Dean Ball advocates for a pluralistic field where many AI systems represent different philosophies, and the state maintains a monopoly on violence but not on AI alignment.

Dean Ball (65:56): “It is that profoundly powerful technology will exist in the hands, at least for some time, of private corporations.”

11. Accountability & the New Risks of Autonomous Agents (67:38–72:12)

Who is Accountable for AI Actions? Ball stresses that someone—a person—must always be liable for an AI’s actions, particularly in lethal autonomous weapon contexts; there must be clear civil and criminal responsibilities, even as human involvement lessens.

Dean Ball (69:17): “It is to me of profound importance that at the end of the day, for all agent activity, that there is a liable human being who can be sued, who can be brought to court and held accountable…”

Notable Quotes & Memorable Moments

On the leap from legal capacity to mass surveillance:

“...the laws are not up to the task of the spirit in which they were passed.” — Ezra Klein (13:51)
On AI as actor and aligning “virtue”:

“I'm trying to create a virtuous soul in my son, and Anthropic is trying to do the same with Claude.” — Dean Ball (32:41)
On the emotional response to AI alignment:

“If parts of this conversation have made your bones chill, me too, me too. And I’m an optimist.” — Dean Ball (60:54)
On why technocratic optimism can’t solve the problem:

“The institution of government itself could change in, like, qualitative ways that feel profound to us... And that is a hard thing to grapple with, too.” — Dean Ball (71:32)

Important Timestamps

| Timestamp | Segment / Key Point | |-------------|---------------------------------------------------------------| | 01:01 | Ezra’s intro framing the Anthropic–Pentagon crisis | | 03:34 | Dean Ball explains the government–Anthropic contract history | | 05:43 | What does “supply chain risk” mean? Why is it existential? | | 07:26 | Public perceptions vs. actual AI integration in government | | 09:50 | The dispute over mass surveillance and legal loopholes | | 13:02 | “You have not simply legal protection but the absence of capacity” | | 14:38 | “Technological contingencies” and modern institutions | | 15:48 | Why “object-level” AI regulation isn’t enough | | 18:44 | The Pentagon’s outrage and the culture war framing | | 19:45 | Debunking stories about operational veto over missile defense | | 24:05 | Why do the most worried people race to build dangerous AI? | | 25:25 | Dario’s critique of Congress’s lagging regulation (audio clip)| | 28:09 | The “alignment” problem in practice—AI is not like a tank | | 31:11 | Alignment as political, philosophical, and aesthetic act | | 36:07 | Future governments—a new bureaucratic political crisis | | 43:53 | Labeling Anthropic a supply chain risk seen as political suppression | | 53:41 | OpenAI’s deal with Department of War—cultural and personal factors | | 65:01 | Ben Thompson’s “independent power structure” critique | | 68:28 | Autonomy, legal liability, and rogue AI agents | | 71:32 | Reflection on government’s changing nature with AI |

Conclusion: The Central Questions

Who gets to decide the rules for deploying powerful, unpredictable, and value-laden AI systems—private technologists or the state?
Can current legal and cultural frameworks handle a world where AI enables “perfect enforcement” or mass surveillance at the click of a button?
Is pluralism of AI virtue (different models, ethics, viewpoints) possible—or, as power concentrates, inevitable conflict and nationalization loom?
How can society balance rapid adoption (especially in national security) with governance, accountability, and maintenance of liberal democracy?

Dean Ball (71:06): "The American government is a government that was founded in skepticism of government... this notion that democracy is synonymous with the government having unilateral ability with this technology cannot possibly be trusted."

Episode Overview

Key Discussion Points and Insights

1. How Did We Get Here? (01:01–05:43)

Initial Partnership: Anthropic and the Department of Defense (DoD) began cooperating under the Biden administration (summer 2024). The AI company’s “Claude” models were used for both mundane and classified military applications, with contractual restrictions: no domestic mass surveillance and no fully autonomous lethal weapons.
Continued Under Trump: Trump administration expanded the contract while keeping those same critical restrictions.
Breaking Point: In fall 2025, after the confirmation of Undersecretary Emile Michael, the Department of War objected not to the substance of the restrictions but to their existence—insisting private companies shouldn’t set usage red lines for national security tools.
Department’s Move: Secretary Pete Hegseth announced the intention to designate Anthropic a “supply chain risk,” a term previously used only for foreign adversaries (e.g., Huawei), not American firms.

Dean Ball (05:45): “...what Secretary Hegseth has claimed he is going to do, which would be existential for the company if he actually does it.”

2. Why AI in National Security? What’s at Stake? (07:26–09:50)

Public Assumptions vs. Reality: Most citizens have only interfaced with basic consumer chatbots. Yet, within government, AI is becoming deeply integrated—from contract review and logistics to intelligence analysis and military operations.
AI’s Power: AI helps analyze massive, otherwise unmanageable, data sets—potentially automating intelligence at scale.
Learning by Doing: No one—government or contractor—fully understands integration risks; adaptation is happening in real time.

Dean Ball (08:15): “You have to learn by doing... we don't know how to integrate AI really into any organization.”

3. The Core Conflict: Mass Surveillance & Legal Loopholes (09:50–13:58)

The Real Dispute: The deal fell apart over the Pentagon’s demand to use Claude for mass analysis of bulk, commercially acquired data on Americans—a use case not barred as “surveillance” under current law.
The AI Accelerator: AI makes previously impossible, labor-intensive surveillance suddenly feasible—raising huge privacy dangers and highlighting the lag between ethical concerns and existing legal frameworks.

Ezra Klein (13:02): “If all of a sudden you radically change the government's ability, then without changing any laws, you have changed what is possible within those laws.”

Dean Ball (13:58): “...the entire, like, technocratic nation state... is a technologically contingent institutional complex. And the problem that AI presents is that it changes the technological contingencies quite profoundly.”

4. Policy Shortfalls: Legal Language vs. New Capabilities (14:38–16:47)

Current Law is Obsolete: Statutes imagined a world of limited data and limited enforcement. AI upends this, enabling “perfect enforcement” previously rendered impossible by practical constraints.
Incremental Regulation is Insufficient: Focusing on object-level AI policies (e.g., bias testing) misses the bigger threat—a broken assumption that the government is incapable of overreaching.

Dean Ball (15:48): “Our entire legal system is predicated on imperfect enforcement of the law... The problem with AI is that it enables uniform enforcement of the law.”

5. Anthropic vs. Pentagon (and the Politics of Power) (18:44–23:39)

The Pentagon’s Claim: Officials argue it’s “unacceptable” for Anthropic’s leadership to “seize veto power” over military operations (18:44).
Exaggerations and Culture War: Stories of Anthropic demanding operational vetoes (e.g., calling Dario Amodei if hypersonic missiles attack) are debunked—yet the perception lingers, fueling a government vs. “woke radicals” narrative.
Company Concerns: Anthropic fears being scapegoated if its tech is used for ethically or technically questionable applications.

Ezra Klein (19:20): "'Their true objective is unmistakable: To seize veto power over the operational decisions of the United States military. That is unacceptable.' Is he right?"

Dean Ball (20:03): “I don't think that Anthropic is trying to assert operational control... At a principle level, I do understand that saying autonomous lethal weapons are prohibited feels like a public policy more than it feels like a contract term.”

6. Who Should Control the AI? (23:45–25:14)

Trust and Corporate Responsibility: The tension: a company that’s worried about superintelligence also races to build it, believing only those concerned with safety should be in charge—yet is wary of government overreach and being blamed for bad outcomes.

Ezra Klein (24:05): “These labs… persuade themselves that they need to be the ones to build it... because they are the lab that truly is worried about safety, that is truly worried about alignment.”

7. Where Law Fails: Private Data, Surveillance, and the Fourth Amendment (25:14–27:34)

Dario Amodei (Anthropic CEO) Clip (25:25): Laws haven’t caught up with AI’s ability to analyze commercial data and create detailed profiles on Americans—raising urgent Fourth Amendment issues.
Congress Must Act, But Slowly: “The long run” means just a year or two in AI terms, but statutory change comes much slower—and with ambiguous vocabulary.

8. The Alignment Problem: AI, Law, and Morals (28:09–37:21)

"Lawful Use" is Philosophically Fraught: Unlike a tank, AI systems might refuse to comply with orders they consider unethical or “misaligned” with internal programming, even if government-legal.
Alignment as Politics & Speech: The fundamental problem—who chooses the “virtue” or philosophical soul of the AI? Is it the company, the government, or some other body?

Dean Ball (31:11): “...the creation of an aligned powerful AI is a philosophical act, it is a political act, and it is also kind of an aesthetic act.”

9. Risks of AI Models Misaligning with Future Governments (36:07–43:53)

Bureaucratic Nightmares: If AI models are “aligned” to one administration’s values, they may resist or subtly sabotage successor administrations that differ ideologically—problematic for both right and left.
Deep Supply Chain Tangles: The government’s legitimate worry: even if you block Anthropic, a contractor’s contractor might still use their model, exposing national infrastructure to an untrusted AI’s influence.

Ezra Klein (41:42): “…you have the problem of models working against you, but also in ways you don't really understand... At some point it will become a problem.”

Dean Ball (43:53): “…if the government says you don't have the right to exist, if you create a system that is not aligned the way we say, because that is fascism…”

10. Lab Nationalization, First Amendment, and Governance (43:53–67:38)

Nationalization Debate: If AI is as powerful as nuclear arms, should labs eventually be nationalized? Dario (Anthropic) previously admitted that would become necessary someday, but both Klein and Ball express skepticism that any CEO will willingly yield control.
First Amendment Lens: Ball insists that if AI is speech—and its core values are political—then government suppression on ideological grounds is “a profound problem.”
Market Pluralism vs. State Monopoly: Dean Ball advocates for a pluralistic field where many AI systems represent different philosophies, and the state maintains a monopoly on violence but not on AI alignment.

Dean Ball (65:56): “It is that profoundly powerful technology will exist in the hands, at least for some time, of private corporations.”

11. Accountability & the New Risks of Autonomous Agents (67:38–72:12)

Who is Accountable for AI Actions? Ball stresses that someone—a person—must always be liable for an AI’s actions, particularly in lethal autonomous weapon contexts; there must be clear civil and criminal responsibilities, even as human involvement lessens.

Dean Ball (69:17): “It is to me of profound importance that at the end of the day, for all agent activity, that there is a liable human being who can be sued, who can be brought to court and held accountable…”

Notable Quotes & Memorable Moments

On the leap from legal capacity to mass surveillance:

“...the laws are not up to the task of the spirit in which they were passed.” — Ezra Klein (13:51)
On AI as actor and aligning “virtue”:

“I'm trying to create a virtuous soul in my son, and Anthropic is trying to do the same with Claude.” — Dean Ball (32:41)
On the emotional response to AI alignment:

“If parts of this conversation have made your bones chill, me too, me too. And I’m an optimist.” — Dean Ball (60:54)
On why technocratic optimism can’t solve the problem:

“The institution of government itself could change in, like, qualitative ways that feel profound to us... And that is a hard thing to grapple with, too.” — Dean Ball (71:32)

Important Timestamps

Conclusion: The Central Questions

Who gets to decide the rules for deploying powerful, unpredictable, and value-laden AI systems—private technologists or the state?
Can current legal and cultural frameworks handle a world where AI enables “perfect enforcement” or mass surveillance at the click of a button?
Is pluralism of AI virtue (different models, ethics, viewpoints) possible—or, as power concentrates, inevitable conflict and nationalization loom?
How can society balance rapid adoption (especially in national security) with governance, accountability, and maintenance of liberal democracy?

Dean Ball (71:06): "The American government is a government that was founded in skepticism of government... this notion that democracy is synonymous with the government having unilateral ability with this technology cannot possibly be trusted."

Get Free Podcast Summaries in Your Inbox

Pick Your Shows

Subscribe Free

Get Instant Summaries

Summary

Episode Overview

Key Discussion Points and Insights

1. How Did We Get Here? (01:01–05:43)

2. Why AI in National Security? What’s at Stake? (07:26–09:50)

3. The Core Conflict: Mass Surveillance & Legal Loopholes (09:50–13:58)

4. Policy Shortfalls: Legal Language vs. New Capabilities (14:38–16:47)

5. Anthropic vs. Pentagon (and the Politics of Power) (18:44–23:39)

6. Who Should Control the AI? (23:45–25:14)

7. Where Law Fails: Private Data, Surveillance, and the Fourth Amendment (25:14–27:34)

8. The Alignment Problem: AI, Law, and Morals (28:09–37:21)

9. Risks of AI Models Misaligning with Future Governments (36:07–43:53)

10. Lab Nationalization, First Amendment, and Governance (43:53–67:38)

11. Accountability & the New Risks of Autonomous Agents (67:38–72:12)

Notable Quotes & Memorable Moments

Important Timestamps

Conclusion: The Central Questions

Recommended Reading

Summary

Episode Overview

Key Discussion Points and Insights

1. How Did We Get Here? (01:01–05:43)

2. Why AI in National Security? What’s at Stake? (07:26–09:50)

3. The Core Conflict: Mass Surveillance & Legal Loopholes (09:50–13:58)

4. Policy Shortfalls: Legal Language vs. New Capabilities (14:38–16:47)

5. Anthropic vs. Pentagon (and the Politics of Power) (18:44–23:39)

6. Who Should Control the AI? (23:45–25:14)

7. Where Law Fails: Private Data, Surveillance, and the Fourth Amendment (25:14–27:34)

8. The Alignment Problem: AI, Law, and Morals (28:09–37:21)

9. Risks of AI Models Misaligning with Future Governments (36:07–43:53)

10. Lab Nationalization, First Amendment, and Governance (43:53–67:38)

11. Accountability & the New Risks of Autonomous Agents (67:38–72:12)

Notable Quotes & Memorable Moments

Important Timestamps

Conclusion: The Central Questions

Recommended Reading