Office of Advanced Artificial Intelligence

Overview

The Office of Advanced Artificial Intelligence (OAAI) is the Department’s center of expertise on frontier computational systems. Its work begins from a narrow question with broad consequences: as automated systems take on control functions in energy, communications, finance and logistics, what happens to the resilience of those systems when the automation behaves unexpectedly, fails, or has to be withdrawn?

The Office leads the Frontier System Evaluation Program, which tests systems not only for what they can do but for how they fail, and in the context of the infrastructure that relies on them. Its evaluation methods are set out in DSC-TR-2026-007. OAAI regards dependence, rather than capability alone, as the variable that most needs measuring: a moderately capable system on which many operators rely, with no tested fallback, can present greater systemic risk than a more capable system used in isolation.

OAAI also provides computational modeling support to the rest of the Department. It maintains the Computational Systems and AI module of the Integrated Civilizational Systems Model, contributed to the Civilizational Systems Outlook, and supports the Neural Continuity Research Program with machine-learning segmentation of connectomic data, simulation methods and independent evaluation of emulation benchmarks. In that last role the Office acts deliberately as an outside check on DNC’s results.

The Office does not regulate, certify or procure AI systems for other agencies. It is a research organization whose findings are shared with operators and with federal science agencies, and published in summary form where doing so does not create exploitable detail. Its evaluation work is conducted in isolated computing environments on ALPHA-1 at Facility Alpha.

Responsibilities

• Lead the Frontier System Evaluation Program and maintain the Department’s evaluation framework for frontier computational systems.
• Map the dependence of critical infrastructure control functions on automated and AI-based systems, including fallback capacity.
• Assess correlated failure risk arising from shared model architectures, training lineage, vendors and compute supply.
• Maintain the Computational Systems and AI module of ICSM and supply evaluated parameters for it.
• Provide machine-learning, simulation and evaluation support to the Neural Continuity Research Program.
• Conduct horizon scanning on emerging computational technologies and their implications for systemic risk.
• Advise the Secretary and Chief Scientist on technological risk matters.
• Share findings concerning specific deployed systems with responsible operators before publication.

Organization

Frontier Systems Evaluation Branch

Designs and conducts evaluation campaigns of frontier systems, including robustness, distribution-shift and degraded-input testing, in isolated environments at Facility Alpha.

Infrastructure Dependency Analysis Branch

Maps where automated decision systems sit in the control paths of critical infrastructure and assesses the capacity of operators to revert to manual or alternative control.

Computational Modeling Support Branch

Maintains the ICSM Computational Systems and AI module and provides segmentation, simulation and machine-learning methods to OSRA and DNC.

Technology Risk Assessment Branch

Conducts horizon scanning and structured assessment of emerging computational technologies, including changes in compute supply and hardware concentration.

Evaluation Integrity Office

Maintains test-set security, contamination controls and reproducibility standards so that evaluation results remain trustworthy and comparable across system generations.