← back
Government Agents: AI Agents Meet Tough Regulations — Mark Myshatyn, Los Alamos National Lab
Original: Government Agents: AI Agents Meet Tough Regulations — Mark Myshatyn, Los Alamos National Lab
Takeaway
Government AI agents require layered FedRAMP/DoD/NIST compliance plus continuous monitoring — vendor trust is the gating constraint, not model quality.
Summary
- Los Alamos has run applied ML since 1956 (Maniac 1 chess); now uses agentic AI for inertial confinement fusion capsule design — agent reads papers, generates code, runs hydrodynamic simulations on HPC.
- Built Venado supercomputer (2500+ GraceHopper nodes, partnership with Nvidia/HPE) and brought OpenAI models onto classified networks for chem-bio safety work.
- Federal AI governance covered: OMB memoranda M-25-21/22 (April), NIST 853 rev 4 (>1000 controls), FedRAMP, DoD CC SRG impact levels, CNSSI 1253 stacks.
- AI use-case governance is still in 180-day rulemaking; nuclear data is 'born classified' so vendor trust posture (continuous monitoring) matters more than SOC 2.
governmentcompliancefedramp
Original description
Lightning talk given at the 2025 AI Engineer World's Fair. https://www.linkedin.com/in/markmyshatyn/