Government Agents: AI Agents Meet Tough Regulations — Mark Myshatyn, Los Alamos National Lab

AI Development & Application at Los Alamos National Laboratory - Los Alamos National Laboratory has been applying AI and ML for almost 70 years, evolving from early applications like Los Alamos chess to modern generative AI agents [2][3] - The laboratory is leveraging generative AI and agentic approaches to accelerate scientific discovery, particularly in areas like inertial confinement fusion (ICF) capsule design [4][5] - The laboratory emphasizes the importance of writing its own models and pushing the science of AI, while also recognizing the need for partnerships with commercial industry and academia [9][10] - Los Alamos National Laboratory is using AI to enhance its mission, including national security, by improving speed and efficiency in various tasks [8] Partnerships & Collaboration - Los Alamos National Laboratory actively seeks partnerships with commercial industry and academia to advance AI development and application [9][10] - The laboratory has established partnerships with the UC family of schools and frontier labs like OpenAI for collaborative research and development [12] - The laboratory emphasizes the importance of trust and shared responsibility in partnerships, particularly regarding the security and governance of AI tools and services [14][15] Governance & Security Considerations - The US government, including Los Alamos National Laboratory, is developing strategies and plans for AI implementation and governance, guided by OMB memorandums like M-25-21 and M-25-22 [15][26] - The laboratory highlights the critical need for robust cybersecurity measures, including compliance with standards like NIST 800-53 and FedRAMP, to protect sensitive data [21][22] - The laboratory emphasizes the importance of building AI systems with explanability, isolation, and governance in mind, particularly for applications with real-world impacts [31][32][35] - The laboratory requires software bills of materials and detailed information on open-source dependencies and patching plans from its service providers [36] Key Requirements for AI Service Providers - AI service providers should prioritize building for explanability to ensure transparency and accountability in decision-making [31] - AI service providers should build for isolation, considering deployment in environments with limited services, such as DoD Impact Level 5 [33][34] - AI service providers should build for governance, providing software bills of materials and detailed information on open-source dependencies and patching plans [35][36] - AI service providers should maintain speed in updating their federal offerings to align with commercial releases, addressing export compliance laws [37][38]