Pentagon’s Michael on Anthropic Talks, Military Use of AI

Anthropic-Pentagon Standoff: A Detailed Summary

Key Concepts:

• Anthropic: A leading artificial intelligence (AI) safety and research company.
• Department of Defense (DoD): The U.S. federal department responsible for national security and the military.
• Defense Production Act (DPA): A U.S. law allowing the government to compel private companies to prioritize defense orders.
• Autonomous Weapons Systems (AWS): Weapons systems capable of selecting and engaging targets without human intervention (often referred to as “lethal autonomous weapons”).
• Human-in-the-Loop: A design principle ensuring human oversight and control in critical decision-making processes, particularly regarding AWS.
• FISA Act (Foreign Intelligence Surveillance Act): U.S. law governing intelligence gathering activities.
• National Security Act of 1947: U.S. law establishing the modern national security apparatus.
• Warfighter: A term referring to military personnel engaged in combat.

I. The Core Dispute & Impasse

The U.S. Department of Defense (DoD) and Anthropic are locked in a dispute regarding the conditions under which Anthropic’s AI technology can be used by the military. Anthropic is facing a 5:00 PM deadline to accept DoD safeguards or risk being blacklisted from future government contracts. The central issue revolves around Anthropic’s refusal to allow its technology to be used for surveillance of U.S. citizens or for autonomous lethal strikes without human oversight. Anthropic’s CEO issued a statement asserting the company’s position, stating, “These threats do not change our position.”

The Undersecretary of Defense responded strongly, publicly calling Anthropic’s CEO “a liar with a god complex,” highlighting a breakdown in negotiations. He explained that the DoD had been negotiating in good faith for approximately three months and believed they had made significant concessions to address Anthropic’s concerns.

II. DoD Concessions & Anthropic’s Objections

The DoD asserts it offered substantial concessions, including written assurances that all use of Anthropic’s technology would adhere to existing U.S. laws and regulations. Specifically, these include:

• Compliance with Existing Laws: The DoD pledged to abide by the National Security Act of 1947, the FISA Act, and all other applicable laws, acknowledging that mass surveillance of Americans is already illegal.
• Human Oversight in Autonomous Weapons: The DoD committed to maintaining human oversight throughout the development, engagement, and use of any autonomous weapons systems, aligning with a pre-existing DoD directive. The Undersecretary noted Anthropic objected to the inclusion of the phrase “as appropriate” at the end of this commitment.

Despite these concessions, Anthropic unilaterally terminated negotiations, publishing an article outlining their concerns before the deadline. The Undersecretary characterized this action as “not good partner oriented practice.” He emphasized that the substantive demands of Anthropic had been met, making the breakdown in communication particularly surprising, especially given the DoD’s successful negotiations with hundreds of other technology companies.

III. Potential DoD Actions & The Defense Production Act

The DoD is considering multiple courses of action if no agreement is reached by the deadline. These include:

• Invocation of the Defense Production Act (DPA): The possibility of compelling Anthropic to provide its technology for defense purposes is being weighed.
• Supply Chain Risk Designation: Designating Anthropic as a supply chain risk, potentially limiting their ability to work with the DoD. The Undersecretary clarified that these are distinct proposals, and the choice depends on the outcome of the negotiations.

The Secretary of Defense, Pete Hegseth, will make the final decision on how to proceed. The Undersecretary maintained openness to further talks, “so long as they’re in good faith,” but expressed concern that Anthropic had launched a pre-planned public relations campaign.

IV. Broader AI Strategy & Partnerships

The DoD is actively pursuing partnerships with multiple AI companies to diversify its options and leverage the potential of AI for national security.

• Existing Partnerships: The DoD has already signed agreements with Google (for unclassified networks) and ex-CIA (for both classified and unclassified networks).
• Strategic Diversification: The goal is to have multiple AI models available to assess their strengths and weaknesses, enhancing the DoD’s capabilities.

The Undersecretary emphasized the importance of AI for future national security, particularly in responding to emerging threats like drone swarms and hypersonic missiles, where rapid reaction times are critical.

V. The Criticality of Lethal Autonomy & Russia-Ukraine War Lessons

The Undersecretary argued that lethal autonomy is becoming increasingly critical for national security, citing the lessons learned from the Russia-Ukraine war. The proliferation of drone swarms and new weapons systems necessitates faster response capabilities than humans alone can provide. However, he reiterated the DoD’s commitment to maintaining human oversight in all autonomous weapons systems. He stated, “we can’t let any one company stand between us and the warfighter because they don't make the rules. Congress thinks the rules.”

VI. Concerns Regarding Anthropic’s Leadership & Potential Overreach

The Undersecretary expressed concerns about Anthropic’s leadership, citing statements about potentially displacing 70 million American workers and ongoing lawsuits related to content scraping. He questioned whether Anthropic is attempting to impose its own values on the American public in a way that undermines democratic processes. He stated, “You do have to worry about are they taking it too far? Are they trying to do and impose their own views on the American people in an undemocratic way?”

VII. Presidential Involvement & Internal DoD Focus

As of the time of the interview, the President has not been directly involved in the negotiations. The matter has remained internal to the Department of Defense.

Conclusion:

The standoff between Anthropic and the DoD represents a critical juncture in the integration of AI into national security. The dispute highlights the tension between technological innovation, ethical considerations, and the need for responsible AI development. The DoD’s willingness to pursue alternative partnerships and potentially invoke the Defense Production Act underscores the importance it places on securing access to AI capabilities, while Anthropic’s firm stance reflects a commitment to its ethical principles. The outcome of this situation will likely set a precedent for future interactions between the government and AI companies, shaping the future of AI in defense.