Can AI Do Intelligence Analysis? Apparently Not.

This document evaluates an AI-driven multi-agent pipeline designed to apply Structured Analytical Techniques (SAT) for intelligence reporting; when tested on the question of likely Chinese cyber operations in a Taiwan contingency, the system produced a polished but ungrounded intelligence assessment (claiming deniable proxy disruptive attacks by groups like APT41 and coordinated PLA SSF activity) that drew on the model's training data rather than the collected evidence, demonstrating critical failures in grounding, collection planning, and analyst-agent behavior.