A recent report from Ars Technica describes how an AI-generated intelligence assessment falsely claimed that a Chinese ship was carrying nuclear components. The hallucination was convincing enough that US military planners came close to launching an attack or boarding operation, until the error was caught in time.
While the near-miss highlights the dangers of relying on AI for critical decisions, the same report notes that the Pentagon's overall use of AI is actually speeding up. Military leaders appear to view such incidents as cautionary tales rather than reasons to pause deployment, suggesting a growing acceptance of AI's role despite its known failure modes.