Lie detectors already have a reputation problem. The Pentagon now wants to see whether AI can make them better. A new Defense Counterintelligence and Security Agency program called Polygraph Next would pair non-contact physiological monitoring with AI and machine-learning algorithms that help score credibility assessments.
The Pentagon is requesting $6.42 million for fiscal 2027, with planned funding through 2031 adding up to roughly $30.3 million (via MIT Technology Review). The budget still lists the overall cost of the program as continuing, so that figure should not be treated as its final lifetime cost. The system is being developed for personnel vetting and insider-threat detection. DCSA also wants automated scoring, decision aids, centralized storage, and analytics tools that can help refine the software and identify new diagnostic features.
So what exactly would the AI be looking at?
Traditional polygraphs track physiological changes such as breathing, heart rate, blood pressure, and skin conductance while someone answers questions. Polygraph Next would move some of that sensing away from physical contact. AI and machine-learning algorithms would then analyze the collected signals and help generate automated scores.
DCSA says non-contact monitoring could reduce inconsistencies caused by sensor setup and speed up assessments. Cloud-based storage and analytics are also planned to help researchers refine the software, address gaps, and study additional diagnostic features.
The idea of letting software hunt for signs of deception is not new. We previously covered VeriPol, a machine-learning system used by Spanish police to flag potentially false robbery reports by looking for patterns in written statements. Then there is Nemesysco, which originally developed its voice-analysis technology as a lie detector for security forces before expanding it into emotion analysis that uses machine learning to find subtle patterns in voice data.
There is still a pretty big scientific problem
Polygraphs do not detect lies directly. They record physiological responses and use those signals to make inferences about whether someone may be deceptive. A landmark 2003 National Academies report notes that changes in respiration, heart rate, blood pressure, and skin conductance can accompany deception, but the same kinds of responses can arise for other reasons.
Accuracy therefore remains a major question. A National Academies review found the scientific evidence supporting polygraph accuracy in security screening to be limited and weak, raising concerns about relying heavily on the tests for personnel decisions. AI may make scoring faster or more consistent, but DCSA has not published evidence showing that machine learning can overcome the underlying limits of polygraph testing.
