Core Announcement
The U.S. Department of Defense has proposed a five-year, $30.3 million program called “Polygraph+” (also “Polygraph Next”) in its latest budget request, aiming to modernize credibility assessment technologies using AI and non-contact sensing.
- Total Budget: $30.3 million
- Duration: Fiscal years 2026–2030
- Leading Agency: Defense Counterintelligence and Security Agency (DCSA)
- Status: Budget proposal submitted to Congress; approval pending
- Technical Focus: AI scoring algorithms + standoff sensing capabilities
Technical Approach and Prototype Clues
Polygraph+ centers on two upgrades: AI and machine learning for scoring, and “standoff sensing”—capturing physiological data without attaching devices to subjects.
The standoff approach has precursors: In 2023, the Defense Innovation Unit (DIU) ran an open call and selected two companies for prototype development. Presage Technologies claims to measure heart and respiration rates via standard cameras. Altec Research—a medical sensor firm expanding into non-contact sensing—developed a system whose DIU-released screenshot indicates tracking of head movement, facial skin temperature, and pore activity. Neither company responded to technical inquiries; DIU declined further comment.
Current polygraph practice remains unchanged since the 1920s: examiners compare physiological responses to baseline questions (“Is the sky blue?”) versus target questions (“Have you committed a crime?”) to infer veracity. Despite tens of thousands of federal tests annually, reliability remains disputed. The 2003 National Research Council (NRC) report concluded evidence for efficacy is “weak at best.”
Counterintuitive Data: Accuracy Claims vs. Real-World Risk
A striking discrepancy exists: the American Polygraph Association cites 80%–94% accuracy, while the 2003 NRC report undermines this. Applying even the lower bound (80%) across the DoD’s 2.8 million employees could yield tens of thousands of false accusations.
The core issue is methodological. Experts identify three deception signals—physiological stress, cognitive load, and conscious concealment efforts—but current polygraphs target only one. Kyri Kotsoglou, a legal scholar at Northumbria University, argues combining AI with polygraphy is “the worst of both worlds”: algorithms may discover new data patterns, but cannot verify their link to lying since polygraph outcomes lack ground truth.
Subjectivity compounds the problem. Results vary widely between examiners; minority candidates face higher deception likelihood. respondents can also learn countermeasures, like mentally boosting baseline responses. Erasmus University’s Sophie van der Zee notes the device’s primary effect is deterrent: many confess before testing begins, but only if they believe it works.
Historical Precedents and Adoption Guidance
Recent cross-modal attempts have all stalled: Silent Talker (UK), iBorderCtrl (EU-funded), and AVATAR (U.S. border application) combined video analysis, voice metrics, and movement tracking yet faded quietly.
Who should adopt: Government security offices handling high-clearance vetting; enterprise insider-threat programs with mature risk frameworks.
Who should wait: Private-sector firms without classified clearance access; judicial systems requiring evidentiary admissibility; organizations needing legally defensible accuracy thresholds.
Final Note
After a century of technological pursuit, lie detection remains constrained by the absence of a universal physiological tell. AI may enhance multi-modal analysis, yet cannot overcome the foundational flaw—that polygraphs measure stress, not deception. Their utility will likely endure as psychological leverage rather than factual arbiters.