NSA turns to AI to triage intelligence data faster
Analysts now spend days or weeks sifting signals-intake for anomalies; the agency's cybersecurity chief says AI can cut that dramatically, though he stopped short of naming a specific timeline.
TL;DR
NSA is exploring AI tools to accelerate how analysts triage the enormous volumes of intercepted communications and electronic signals the agency collects daily, cybersecurity directorate chief David Imbordino said Tuesday at the Billington Cybersecurity Summit. "Volume has always been a problem," Imbordino told the Five Eyes panel, framing AI as a way to surface "the nuggets" faster than methods that currently take days or weeks. The effort sits alongside NSA's broader role implementing a June executive order that directs national security agencies to create classified tests for frontier AI models with advanced cyber capabilities.
Imbordino's remarks are the clearest public signal yet that NSA sees AI triage as an operational answer to a problem the agency has lived with for decades: there is simply too much raw signal to process at human speed. The cybersecurity directorate handles both intelligence collection and cyberdefense, and the data flows don't pause while an analyst works through a backlog.
What's new is the confidence. Imbordino didn't describe a pilot or a proof-of-concept. He described a near-future in which analysts "increasingly use AI tools to look at disparate pieces of information and evaluate them much more quickly." That's a deployment posture, not a research posture.
The context matters. In late August, NSA deputy director Tim Kosiba said the agency wants access to "all the models" and is in active discussions with frontier AI companies. The June executive order (which NSA is helping implement) calls for classified tests that evaluate when advanced models have developed hacking capabilities serious enough to warrant additional scrutiny. The triage use case Imbordino described is a different lane: it's about making analysts faster at their existing jobs, not about red-teaming models for offensive cyber risk. But the two efforts draw from the same underlying bet that AI is now mature enough to handle work the agency can't staff its way out of.
Published ·Deep Fathom