The AI Policy Questions in the Underlying Issues of the Anthropic-Pentagon Case

Jennifer Huddleston

The Department of War’s decision to label Anthropic a supply chain risk raises significant concerns related to government power over private business and the First Amendment. The litigation involving the designation is rightly focused on questions of due process and the First Amendment. As Judge Lin notes in the preliminary injunction against the government, disputes over potential government surveillance, autonomous lethal weapons, and the extent of all lawful uses are “not for this Court to answer in this litigation.”

The court is correct. Such questions are clearly within Congress’s authority under Article I of the Constitution. The facts underlying this dispute underscore the need for an appropriate policy framework to support innovation and provide clear guidance to innovators, citizens, and the government for protecting civil rights and civil liberties.

Civil Liberties, Domestic Surveillance, and AI

AI has fantastic potential to analyze large amounts of data better than any human reviewer could. As a technology, such potential gives it potential beneficial use cases across both the public and the private sectors. This enables it to better detect cancer, predict and respond to national disasters, and help Americans start new businesses. But as Uncle Ben says in the 2002 version of Spiderman, “With great power comes great responsibility.” 

Given the potential power of this technology, there are scenarios where their potential impact on civil liberties must be considered. When it comes to government use of these powerful tools, we must ensure that appropriate guardrails are in place to protect civil liberties while minimizing the potential impact on beneficial uses and the overall development of the technology.

I would suggest four common-sense steps could be taken in this regard:

1. Provide clarity and guardrails around what government data can be used for and how that data can be used.
2. Close existing loopholes that allow the government to buy bulk data via data brokers.
3. Engage on policy reforms for existing programs like FISA Section 702 that could be abused or expanded to result in domestic surveillance, particularly in the AI era.
4. Establish appropriate approval processes, such as subpoenas and warrants for use of technology by law enforcement or other agencies, in line with existing Fourth Amendment doctrine and other civil liberties.

None of these restricts the development of AI itself but rather restrains the government in its deployment or provides clarity around existing law in light of new technology to prevent potential abuse. Additionally, these data practices are a wise move beyond just AI technology and would be technology-neutral for both current and future technologies. 

The case for FISA reform and other surveillance-related protections has been made well elsewhere by my colleague Patrick Eddington. While this issue is particularly important given the debate around reauthorization, rather than engage in a full discussion here, I would recommend those interested in the topic consider his robust body of work on the topic.

Improving Data Governance and Protection

While an individual can decide whether or not they want to give their data to a particular private company, there are many situations where that is not the case with the government. As a result, particularly given the rise of AI, many are concerned about the way data could be abused by bad actors or the government itself. 

Existing laws on census and statistical data provide examples of what guardrails for the government’s own data could look like. Under CIPSEA, enacted in 2002, federal government data collected for statistical purposes may not be disclosed in a way that would identify an individual without that individual’s consent. There are also protections around various data sets, like Census data and other statistics, that limit their use by the government to the purposes for which they were collected.

Now, government data sets can have positive use cases for technological development, the public, and researchers. The above should not be seen to limit such availability. Instead, such guardrails could help ensure that data collected by one agency for a legitimate purpose cannot be abused by another agency for an unintended purpose. At the same time, this can still allow the use of such data in the public interest, including by those developing AI technologies without access to the resources of large companies.

Clarifying Fourth Amendment Protections in the AI Era

Technology can be incredibly useful, including in law enforcement. However, the use of certain technologies by government actors without appropriate restrictions can raise civil liberty concerns. These concerns are not unique to AI, but they have amplified the need for clear guardrails in the digital age.

Government attempts to circumvent existing restrictions in order to obtain data they would not otherwise have access to predate AI. But the potential impact of AI makes it more important to ensure the loopholes that enable such circumvention are closed. One notable example of this is the “data broker loophole,” which allows large amounts of data not otherwise obtainable by the government to be bought from data brokers. Such acquisition of data goes beyond the scope of a single investigation to include the data of those not at all connected to the underlying law enforcement concern. 

These checks should also be in place for the use of certain technologies that can be helpful but also create additional civil liberty concerns. For example, existing laws require court-approved warrants supported by sufficient probable cause for monitoring emails or conducting wiretaps. Such guardrails lessen the risk of abuse while ensuring law enforcement can access these technologies in appropriate cases. This sort of procedural requirement applied to the data broker loophole would protect civil liberties and help ensure critical evidence is properly obtained and admissible. Several states have taken various steps to respond to such issues regarding law enforcement’s use of facial recognition to ensure a balance. In light of ongoing policy concerns, there should be consideration of appropriate limitations on federal law enforcement that could be codified to protect civil liberties from technological uses that would not be possible in an analog era.

Autonomous Lethal Weapons

For many, AI raises concerns about the potential of killer robots akin to the movie Terminator. As early as 2018, thousands of leading AI scientists called to pre-emptively ban the use of the technology in autonomous lethal weapons or the deployment of weapons of mass destruction. Policy on this topic must balance a range of concerns while being realistic about where technology currently is. 

As Anthropic indicated in its statement on February 26, 2026, the reliability of the technology is such that there are still significant concerns over whether it would be able to power autonomous lethal weapons without putting civilians or even America’s own warfighters at risk. Humans are also fallible, of course. But for such a significant decision, such as the taking of a human life, it is reasonable to expect that an accountable human is ultimately involved in the decision. 

Policymakers have considered this in some defense decisions. The 2025 National Defense Authorization Act (NDAA) included language to preserve “meaningful human control” in nuclear launch decisions.

Policy must not, however, be so broad as to ban the use of AI more generally in defense situations. For example, AI could play roles in alerting potential security breaches, ensuring safe handling and manufacturing through automation, and analyzing potential combat needs. Many technologies like drones or autonomous vehicles rely on AI applications and may be used in defense or combat in the future, like civilian usage. With this in mind, a policy framework to address the legitimate concerns around potential autonomous lethal weapons should focus on the specific potential harm of the decision to use lethal force and the accompanying risk to civilians or others. An overly broad policy could prevent appropriate and even already common uses of automation and AI that are important to maintain readiness and success for defense.

Conclusion

There are significant policy questions raised by the situation between Anthropic and the Pentagon. These policy questions, however, should be appropriately decided not by bureaucrats or the courts, but by Congress. While we are still in the early days of AI, these questions are becoming increasingly pressing for innovators, consumers, and even the government itself.

Source: https://www.cato.org/blog/ai-policy-questions-underlying-issues-anthropic-pentagon-case