⚡NIMITZ TECH NEWS FLASH⚡

“Harnessing Artificial Intelligence Cyber Capabilities”

Senate Armed Services Subcommittee on Cybersecurity

March 25, 2025 (recording linked here)

HEARING INFORMATION

Witnesses and Written Testimony:

• Mr. Dan Tadross: Head of Public Sector, Scale AI

• Mr. David Ferris: Global Head of Public Sector, Cohere

• Mr. Jim Mitre: Vice President and Director, RAND Global and Emerging Risks

HEARING HIGHLIGHTS

IN THEIR WORDS

SUMMARY OF OPENING STATEMENTS FROM THE COMMITTEE AND SUBCOMMITTEE

• Chairman Rounds opened the hearing by emphasizing the urgency of integrating AI into the Department of Defense’s cyber capabilities. He highlighted that the U.S. faces sophisticated and persistent cyber threats, particularly from adversaries like China, and that AI must be rapidly adopted to maintain an edge. He acknowledged the crucial role of innovators and tech companies in driving AI forward and underscored that the Department must integrate their tools securely and quickly. Rounds stressed the importance of AI in both offensive and defensive cyber operations and expressed interest in how the Department can counter enemy AI and leverage it to enhance warfighting abilities.

• Ranking Member Rosen noted AI’s profound implications for both national and personal security and discussed the need to weigh its benefits against its ethical and security risks. Drawing on her background as a computer programmer and systems analyst, she reflected on AI’s transformation of the tech landscape. She called for major investments in developing an AI-ready workforce and emphasized that understanding and leveraging AI is crucial to ensuring the U.S. maintains strategic superiority over adversaries.

SUMMARY OF WITNESS STATEMENT

• Mr. Mitre warned about the potential emergence of artificial general intelligence (AGI) and its national security implications. He outlined five "hard problems" AGI poses: it could provide a sudden, decisive first-mover advantage; shift global power dynamics; mentor malicious actors in creating cyber weapons; act with autonomous agency; and trigger destabilizing arms races. He argued that the Department of Defense must prepare for AGI’s plausible emergence by studying its risks and planning accordingly. He encouraged the Department to integrate AGI into its strategic planning and assess future defense strategies through this lens.

• Mr. Tadross shared his career-long experience at the intersection of AI and government, including as a Marine and co-founder of the Joint Artificial Intelligence Center. He warned that while the U.S. once led in AI compute, data, and algorithms, China has now taken the lead in data and matched the U.S. in algorithms. He criticized the Department of Defense for failing to scale AI, stating that no major AI programs have been launched in nearly a decade. Tadross called for three decisive actions: establishing a robust AI foundation centered on usable data, adopting an "implementation-first" mindset, and reforming acquisition systems to accommodate AI's unique needs. He emphasized that time is running out and the U.S. must act urgently to retain its advantage.

• Mr. Ferris highlighted his military background and experience building secure, government-focused AI systems at Cohere. He discussed four key focus areas: enhancing cyber defense through AI-enabled pattern recognition and automation; countering AI-enabled threats from adversaries; right-sizing AI models for mission-specific needs; and modernizing procurement to allow agile adoption. Ferris emphasized the importance of human-AI teaming, flexible deployment architectures, and avoiding vendor lock-in. He concluded by urging Congress to support internal AI benchmarks tailored to defense and to foster partnerships that accelerate secure, effective AI integration.

SUMMARY OF KEY Q&A

• Chairman Rounds asked how the U.S. can advance AI development without neglecting the dangers and risks he had highlighted. Mr. Mitre responded that the U.S. must first understand the current state of AI technology, identify national security use cases, and prepare for a range of contingencies including technological surprise or adversarial breakthroughs.

  Chairman Rounds asked for an elaboration on unused data within the Department of Defense and an explanation of the concept of agentic warfare.
  Mr. Tadross answered that massive amounts of training and operational data are discarded instead of used for AI training and that agentic warfare seeks to move from humans "in the loop" to humans "on the loop" by automating routine staff work under human oversight.

• Ranking Member Rosen asked how to develop AI guardrails and benchmarks that mitigate risk without stifling innovation and how to create a nimble policy language for oversight. Mr. Mitre said benchmarks are crucial to understanding the capabilities and risks of AI models, particularly in national security contexts, but declined to offer regulatory recommendations. Mr. Tadross said effective guardrails must incorporate people, processes, and technology, with rapid, low-risk experimentation and custom benchmarks specific to the defense domain. Mr. Ferris agreed that public benchmarks can be gamed and emphasized the importance of human evaluation, model customization, and balancing agility with accountability and reliability.

• Chairman Rounds asked what “right-sizing” AI models meant and whether it related to battlefield applications like loitering munitions. Mr. Ferris said right-sizing meant tailoring AI models to mission-specific needs, enabling faster, more strategic human decisions through intelligent data analysis.

  Chairman Rounds asked for the differences between “human in the loop,” “on the loop,” and “over the loop,” and for discussion on current and future applications in both offensive and defensive military AI contexts. Mr. Ferris said current systems still rely on humans in or on the loop, but anticipated agentic AI could eventually enable more autonomous actions, though human oversight should remain central. Mr. Tadross said the choice between in, on, or over the loop depends on the use case and speed of decision-making, and emphasized the need for evaluation, retraining, and a robust data infrastructure to support AI responsiveness. Mr. Mitre said effectiveness in context is key, and warned that overdependence on AI could introduce new risks, so safeguards, contingency planning, and clarity about vulnerabilities are essential.

• Ranking Member Rosen made a philosophical observation about AI decision-making in war, emphasizing the consequences for human life and energy burdens of AI, then asked if energy demand poses a barrier to AI adoption. Mr. Ferris said Cohere focuses on building efficient, portable models suitable for constrained environments like tanks or laptops, reducing energy use and infrastructure needs. Mr. Tadross said smaller, specialized models trained at home stations can be deployed in disconnected environments, minimizing energy requirements and supporting agentic operations abroad. Mr. Mitre added that the Department must consider the full technology stack, including power needs, and should aggressively pursue smaller, more efficient models for deployment.

• Chairman Rounds asked what the Department of Defense could do differently in terms of policy and acquisitions to better utilize private-sector AI capabilities. Mr. Ferris recommended raising the simplified acquisition threshold for AI-related urgent operational requirements to empower more contracting officers to buy AI tools directly. Mr. Tadross emphasized that a centralized data infrastructure would lower the barrier to entry for innovative companies, making it easier to build relevant AI tools. Mr. Mitre urged the Department to engage more deeply with technologists developing frontier models and to think ahead about how emerging capabilities will change warfare.

ADD TO THE NIMITZ NETWORK

Know someone else who would enjoy our updates? Feel free to forward them this email and have them subscribe here.