The Pulse
Pentagon Asks $6.4 Million to Start an AI-Scored Polygraph
The Defense Counterintelligence and Security Agency’s fiscal 2027 budget request asks for $6.421 million to begin Polygraph Next, a program to modernize polygraph testing with non-contact sensors and AI-based scoring, and projects $30.342 m
AI.info Team ·
The Pentagon’s plan for an AI-assisted polygraph promises more accurate credibility assessments. A National Research Council review, by contrast, concluded that physiological responses measured by polygraphs are not uniquely tied to deception and that the evidence did not justify relying on the tests to screen federal employees. The Defense Counterintelligence and Security Agency (DCSA) is now asking for $6.421 million in fiscal 2027 to begin developing a new system that combines non-contact sensors with AI-based scoring, and projects $30.342 million for it through 2031.
$6.421 million to start, $30.342 million projected
The project, called Polygraph Next, appears in DCSA’s fiscal 2027 research and development budget justification. Its projected spending totals $30.342 million across fiscal years 2027 through 2031: $6.421 million in 2027, followed by annual amounts of $6.449 million, $6.158 million, $5.660 million and $5.654 million. Only the first of those figures is the fiscal 2027 request; the rest are projections, and the document lists the program’s total cost and cost to complete as “Continuing”. None of it is money already spent or a final congressional appropriation.
DCSA describes a system intended to modernize polygraph and other “credibility assessment” technologies. The agency plans to test non-contact physiological sensors, automated scoring and decision-support tools, alongside centralized data storage and analytics. Its stated uses include personnel vetting and insider-threat detection.
Tests are planned before performance is established
The budget document outlines work still to be done, not a validated product. DCSA says it will conduct independent validation studies, field testing and prototype evaluations; its fiscal 2027 plan includes operational pilots with interagency polygraph programs. The agency also plans to develop cloud-based storage and analytics tools to refine software and investigate possible new diagnostic features.
That distinction matters: the budget describes a goal of faster, more accurate assessments, but does not present results showing that the proposed sensors or AI scoring can reliably identify deception. Nor does the document specify a vendor, a final system design or a measured accuracy target. The proposal commits to testing performance across different operational applications and polygraph formats.
A long-standing dispute over physiological signals
The National Research Council’s 2003 report, The Polygraph and Lie Detection, examined the science behind polygraph use, including federal employee screening. Its committee found that the evidence base was limited and that responses measured by the instruments could arise from processes other than deception. Presenting the findings, the committee’s chair, Stephen E. Fienberg, a professor of statistics and computer science at Carnegie Mellon University, put its policy conclusion plainly:
“Our main policy conclusion is that the polygraph's accuracy in distinguishing actual or potential security violators from innocent test takers is insufficient to justify the federal government's reliance on this method for employee screening.”
Stephen E. Fienberg, chair of the National Research Council committee on the polygraph, 8 October 2002
That review predates Polygraph Next and does not evaluate its proposed non-contact sensors or AI scoring. It does, however, set out the scientific problem the new program must address: physiological signals can have multiple causes, so a system’s credibility claims require careful validation in the conditions where officials plan to use it.
The schedule points to pilots, not deployment
The White House submitted its fiscal 2027 budget request to Congress on April 3, 2026. DCSA’s justification projects a development schedule for Polygraph Next running from the first quarter of fiscal 2028 through the third quarter of fiscal 2029, while listing operational pilots and further software and analytics development among its plans. Those milestones describe proposed work; they do not establish that the system has passed testing or will be adopted.
For now, the concrete commitment is a request for staged development funding and a plan to test prototypes. Whether the approach improves credibility assessments will depend on the validation studies and field results DCSA says it intends to conduct.