Intro. [Recording date: March 12, 2026.]
Russ Roberts: Today is March 12th, 2026, and I’m pleased to welcome Dean Ball. Dean is a Senior Fellow at the Foundation for American Innovation, a Policy Fellow at Fathom, and the author of the AI-focused newsletter Hyperdimensional, available on Substack. His expertise lies in technological change, institutional evolution, and the future of governance. Prior to his current roles, he served as Senior Policy Adviser for AI and Emerging Technology at the White House Office of Science and Technology Policy, where he played a pivotal role in drafting America’s AI Action Plan.
Dean, welcome to EconTalk.
Dean Ball: Thank you for having me, Russ.
Russ Roberts: Today, we’ll delve into the intriguing relationship between private companies engaged in artificial intelligence, such as Anthropic, which developed the LLM known as Claude, and the Department of War. We’re particularly focused on the recent conflict regarding the regulations governing Claude’s military applications. This situation has sparked a significant debate over the constitutional ramifications of private sector and federal government interactions, which we will unpack today. Our discussion draws on your insightful article from your Substack, Hyperdimensional, aptly titled “Clawed.” Let’s start at the beginning: what triggered this conflict, and what are the key issues at play?
Dean Ball: To fully grasp this conflict, we should rewind about 18 months to late in the Biden Administration. In the summer of 2024, the Department of Defense (now the Department of War) approached Anthropic to establish a contract for using the Claude model in classified applications. This was a notable distinction from unclassified uses. The Department of Defense, along with other government agencies, frequently employs LLMs for mundane tasks like reviewing contracts or navigating HR protocols, but this contract involved more sensitive applications, such as intelligence analysis and operational targeting in combat zones.
The contract included usage restrictions, particularly prohibiting domestic mass surveillance and the deployment of AI in autonomous lethal weapons—essentially, machines that could identify, track, and eliminate targets without human intervention. These restrictions were agreed upon by the Department of Defense.
Fast forward to the summer of 2025, under the Trump Administration, the Department of Defense expanded this contract significantly—up to $200 million—while renewing its terms, including the same restrictions on mass surveillance and autonomous lethal weapons.
Then, in the fall of 2025, a newly confirmed Department of War official, Emil Michael, reviewed the contract and asserted that the Department could not adhere to these restrictions. He proposed renegotiating for “all lawful use.” Anthropic was willing to drop several restrictions but stood firm on the two key issues, which became a red line for them. The Department of War escalated the situation, threatening to label Anthropic as a supply chain risk, which would not only cancel their Department of War contracts but also jeopardize their contracts with other Department of War contractors, such as Microsoft.
This conflict unfolded over several months, culminating in Anthropic filing a lawsuit against the government in California’s Northern District. And here we are today.
Russ Roberts: To clarify, many Americans would likely oppose the Department of War conducting mass surveillance. However, what precisely constitutes mass surveillance? Would the federal government need a court order for certain surveillance activities? It seems the Department of War is seeking the ability to conduct mass surveillance that is deemed “legal.” This could include actions that many would classify as mass surveillance, alongside the use of approved autonomous lethal weapons. Anthropic appears to want a contractual distinction that allows them to reject uses of their technology that they find ethically objectionable, even if those uses are technically legal. Is that an accurate summary of their stance?
Dean Ball: Yes, that’s correct. The definition of domestic mass surveillance is indeed a complex sticking point here. For instance, a plethora of commercially available data sets can include sensitive information about Americans, such as smartphone location data. Many apps, like weather applications, require location access, and companies often monetize this data. There’s also commercial satellite data and web usage data that can be aggregated for insights on individuals.
This has been a long-standing reality in the era of web-scale data. However, the constraint lies in the time and resources needed to analyze this data for individual targets. With advanced AI systems, the intelligence community could theoretically scale its analysis capabilities infinitely, transforming a workforce of thousands into millions of analysts.
Anthropic argues, and I concur, that the law has not adequately adapted to this new reality. The legal definition of domestic mass surveillance doesn’t align with the everyday understanding of the term. This discrepancy complicates the conversation significantly.
Russ Roberts: So, what are the stakes in this situation? As we record this in March 2026, things could shift dramatically by the time this airs. So, what’s at risk here? It’s worth noting that after the dispute with Anthropic, the Department of War has allegedly reached an agreement with OpenAI that carries similar terms but lacks the same restrictions. Is that accurate?
Dean Ball: Yes, that’s right. There appears to be an agreement in principle for using OpenAI models in classified settings, but without the same red-line protections Anthropic sought. OpenAI’s approach emphasizes technical safeguards; they plan to control the deployment of their systems to the Department of War, allowing the AI to assess its actions in real-time regarding domestic mass surveillance.
Russ Roberts: We’ll see how that unfolds. Why do you find this alarming? What are your concerns about the actions of the Department of War?
Dean Ball: There are several reasons. First, the concept of “all lawful use” is problematic. Defense procurement experts I’ve spoken with consider it an unusual contracting notion, as it raises questions about who defines what is lawful. In this case, the Trump Administration is essentially claiming, “We decide what lawful is until someone stops us.”
Moreover, the nature of the punishments threatened against Anthropic is concerning. Secretary of War Pete Hegseth warned that regulations could prevent any Department of War contractor from doing business with Anthropic. This is a significant escalation, as it impacts commercial relationships beyond just Department of War contracts.
The actual supply chain risk designation that was ultimately issued means that while Anthropic can still operate commercially, they cannot be utilized in fulfilling Department of War contracts. Microsoft, for instance, may use Claude for other applications, but not in connection with any Department of War project.
Russ Roberts: Additionally, there are rumors that Anthropic’s more safety-oriented culture and its perceived political leanings are contributing factors in this dispute. Do you think that’s a relevant point?
Dean Ball: Yes, that’s worth mentioning. The supply chain risk designation is often reserved for companies seen as foreign adversaries, implying a level of hostility toward Anthropic. It’s treating Anthropic as if they are enemies of the state.
Russ Roberts: If the broader designation had been enacted, it would have equated Anthropic with a terrorist organization, further emphasizing the stakes involved.
Dean Ball: Exactly. This situation is just the beginning. If the Administration decides to mobilize the entire federal regulatory apparatus against Anthropic, they certainly can. Additionally, Anthropic has alleged that government officials are encouraging their customers to sever ties with the company, which complicates matters further.
In summary, the government is essentially stating that if you do not comply with the terms they unilaterally establish, they will take punitive actions against your company. This is a clear encroachment on private property rights.
Russ Roberts: How might the opposing side frame their argument? They could argue that the government must ensure national security and that Anthropic is obstructing this imperative. How would you summarize their perspective?
Dean Ball: The opposing argument would claim that Anthropic is overstepping its bounds by using its private power to influence public policy. They could argue that the government should be trusted to determine what is legal and appropriate for national security. Essentially, they might say that as long as the government is acting within the law, there shouldn’t be restrictions from a private company.
Russ Roberts: So, the argument boils down to the assertion that private enterprises should not wield the power to dictate public policy through their contracts.
Dean Ball: Correct. The government would contend that public policy cannot be dictated by a private entity, regardless of the competitive environment. Additionally, there are allegations that Anthropic threatened to withdraw Claude’s services during military operations if they disagreed with the government’s actions, which complicates the narrative.
Russ Roberts: The analogy of a private company withholding a nuclear weapon from the government based on ethical concerns is an interesting point. How does that analogy apply here?
Dean Ball: The contractual analogy is valid. One could envision Anthropic requiring terms that limit their models’ applications to wars declared by Congress. The point remains that while the government has a fair argument, they could address their concerns without attempting to dismantle Anthropic’s business.
Moreover, if these protections matter so significantly to Anthropic, they should have insisted on legal safeguards before entering into a contract.
Russ Roberts: It’s essential to recognize that the U.S. has a peculiar relationship with private companies involved in national defense. Companies like Boeing or McDonnell Douglas operate in a hybrid space, reliant on government contracts while maintaining private status. This dynamic complicates the conversation about AI technology, which, while primarily civilian, holds substantial military potential.
Dean Ball: Indeed, we’re grappling with the implications of AI’s dual-use nature. It has the potential for both civilian and military applications, raising questions about regulation and oversight. While some argue for stringent government control, it’s crucial to consider how these technologies can also empower individual liberty and facilitate knowledge acquisition.
Russ Roberts: The ongoing evolution of government regulation amidst rising executive power poses significant challenges. As you’ve pointed out, the stakes are high, particularly in the realm of emerging technologies like AI. The balance between public interest and private innovation is precarious.
As we navigate this landscape, it’s vital to understand that the government’s heavy-handed approach may stifle innovation and risk eroding fundamental principles of private property rights and freedom of expression.
Dean Ball: Absolutely. It’s a nuanced debate, and we find ourselves at a crossroads. The future of AI governance is uncertain, and as we consider the implications of unchecked executive power, we must also explore the role of private corporations in shaping our technological landscape.
Russ Roberts: You began your piece with a poignant personal story about your father’s passing. How does this relate to the current moment in American governance?
Dean Ball: My reflections on my father’s death made me realize that institutions, much like living beings, undergo processes of erosion. Over the years, I’ve witnessed the gradual decay of core principles that underpin our Republic. This erosion feels akin to the slow death of my father, where the essence of what once was gradually fades away.
It’s painful to acknowledge this decline, but it’s a reality that compels us to confront the implications of our current trajectory.
Russ Roberts: Your insights resonate deeply, particularly in the context of our current political climate. Many observers might attribute these issues to specific administrations, but as you pointed out, this is a broader trend that transcends individual leaders.
The gradual erosion of constitutional constraints on government power raises serious concerns about the future of our democratic institutions. It’s imperative that we recognize these patterns and engage in meaningful discussions about the role of government and the preservation of our Republic.
Dean Ball: I completely agree. While elections may still occur, the underlying principles of our Republic are in jeopardy. Recognizing the chilling effect of unchecked executive power is essential for safeguarding democracy.
Russ Roberts: The power dynamics between the Executive Branch and the Courts will likely become increasingly contentious. The current administration’s approach to governance may challenge the foundational principles of our democratic system.
Dean Ball: Yes, the Courts are currently the last bastion against unfettered executive power, but their effectiveness is contingent on the willingness of the Executive to observe their rulings. As we navigate this complex landscape, we must remain vigilant in our commitment to preserving the rule of law and the principles that underpin our democracy.
