- Defense Counterintelligence and Security Agency (DCSA) says its Modernizing Polygraph effort excludes large language models, facial coding and vocal analysis, though it did not address AI deceptive-speech models shown in earlier slides.
- Experts warn AI deception detection lacks scientific backing, with research showing roughly 50:50 accuracy and a tendency to wrongly assume dishonesty.
- Critics also raise bias, automation bias and due process concerns, as budget documents price the Polygraph+ project at $31 million.
Has the Pentagon given up on AI polygraph analysis of trustworthiness?

Loading comments…

