AI is rapidly impacting all sectors of society, but leaders in the intelligence community in particular find themselves grappling with the ever-evolving question of how to safely reap its benefits for their own work. In its second panel session, the NSDPI 2026 Annual Conference explored key tradecraft implications of this new technological era, with NSDPI Senior Fellow Brian Lessenberry leading the discussion. A former senior intelligence executive who has held leadership roles at ODNI, DOS, and NGA, Lessenberry began by soliciting some broad perspectives from his panel of academic, government, and industry experts.
Amy McAuliffe, a distinguished visiting professor of the practice at the University of Notre Dame, shared that the analytic lessons the IC learned from Iraq were highly relevant here; factoring in confidence levels, acknowledging gaps in intelligence, and incorporating individual analysis into mainline analysis and predictions are all important. “We do not want rapidity and recency bias” she cautioned, “which AI models are dominated by.” Former IC Senior Executive Wayne McCool acknowledged these concerns but added a note of optimism, saying that AI will help with confidence levels and other important considerations if experts know how to properly integrate it into analysis. McCool shared that much of the IC has been using AI for years, bringing it into workflows related to automated tasks such as map-making and helping analysts digest overwhelming amounts of data.
As conversation shifted to the topic of model design, Rhombus Power Vice President of Intelligence Bruce Frost offered a unique perspective, informed by his current work in private industry and his over thirty years of federal service. “Are you trying to convert a model made for a commercial purpose to serve an intelligence purpose? Or are you building a model from scratch? It’s usually better to build from scratch” he explained. Just as analysts can carry biases they must interrogate, machine learning models can be biased as well. Frost emphasized that companies providing intelligence have major responsibilities as AI continues to reduce data processing timelines and the IC moves more rapidly toward private-public partnerships to reap the productivity benefits.
While these central questions around the safe and efficient incorporation of AI tools into tradecraft are crucial to consider, the perspective of intelligence end consumers should not be forgotten. “I don’t think a lot of policymakers understand or care about the operations of tradecraft,” Rebecca Hersman, a senior research scholar at the Centre for the Governance of AI, reminded the rest of the panel. She explained that many leaders are skeptical following the intelligence failure in Iraq, and when decision makers find themselves frustrated by the opacity of the sources and methods behind IC conclusions, they may turn to their own LLMs as trusted advisors.
As the conference’s AI and Intelligence session wound down, panelists shared some key conclusions about the promise and perils of machine learning models. Hersman highlighted AI’s problematic reluctance to reveal that it doesn’t know an answer in the interest of being helpful, pointing out the danger of losing touch of our unknown unknowns as we assess our confidence in our conclusions. McAuliffe cautioned that, while AI’s ability to explain its decisions and reasoning are crucial to that confidence, that capability does not yet exist in the public or private sector. And finally, McCool emphasized that AI cannot provide values-based judgements, and should not be a part of targeting decisions or party to justification of force questions. Acknowledging these considerations, Frost stressed that keeping humans in the loop and synchronizing all data sources into high quality intelligence products is crucial, and that targeted AI models can be incredibly helpful in this process.
