The Pentagon is asking Congress for $30.3 million to build an AI-powered lie detector over the next five years. The program, called Polygraph+ or Polygraph Next, aims to modernize federal credibility assessments with machine learning and sensors that can take physiological readings without physical contact. Experts who study deception detection call the effort misguided and warn it will compound existing flaws with new uncertainty.
According to a Department of Defense budget document first reported by Inside Defense, the project will "modernize federal polygraph and credibility assessment technologies" to improve their accuracy and reliability. The Defense Counterintelligence and Security Agency (DCSA), which conducts background checks for the federal government, would run the program. Congress has not yet approved the funding.
How the technology would work
Polygraph+ will focus on scoring algorithms that use artificial intelligence and machine learning, along with a technique called "standoff sensing" - the ability to take physiological readings without attaching a device to a subject. The budget document says the new technology will be used for vetting prospective employees and "insider threat detection." The DCSA did not respond to a request for more information about which specific technologies will be used.
Other Pentagon efforts offer potential clues. In 2023, the Defense Innovation Unit (DIU) selected two companies to build deception detection prototypes. Presage Technologies claims to measure heart rate and breathing rate using standard cameras. Altec Research, a medical sensor company branching into non-contact sensing, built a prototype that tracks head movement, facial skin temperature, and pore activity. Neither company responded to requests for comment, and the DIU declined to comment.
A long history of unreliable results
Current polygraph technology has barely changed since the 1920s. Examiners measure blood pressure, pulse, breathing, and sweat to compare physiological responses to baseline questions and target questions. The federal government conducts tens of thousands of these tests each year while screening employees, but the reliability has been repeatedly challenged. Results are rarely admissible in court.
In 1983, Congress's Office of Technology Assessment concluded there was very limited evidence supporting the polygraph's use for employee screening. A 2003 National Research Council report said evidence for its efficacy was "weak at best." Research suggests humans can spot a lie just over half the time without technical assistance. The American Polygraph Association claims 80% to 94% accuracy, but the NRC report pointed out that a screening test at that level would still produce tens of thousands of false accusations when applied across the DOD's 2.8 million employees.
Polygraph interpretations are often subjective. Different examiners get wildly different results, and people from minority groups are more likely to be judged as deceptive. With training, subjects can also learn countermeasures - like stepping on a pin hidden in a shoe to artificially heighten their physiological response to baseline questions.
"If you know how it works, you can beat it," said Sophie van der Zee, an associate professor who studies deception at Erasmus University in Rotterdam. She said the machine's biggest effect is deterrence. "But that only works if people think a polygraph works."
Why AI won't fix the fundamental problem
Various new strategies for lie detection have been attempted over the decades - thermal cameras, pupil trackers, brain scans. None have yielded reliable results outside the lab. The problem is that there is no single telltale sign of lying that holds true for everyone all the time. "There is still no Pinocchio's nose," van der Zee said.
AI could theoretically improve polygraphs by finding patterns in data that examiners miss. Algorithms are more likely to be used for "multi-modal" deception detection, combining multiple measurements into an overall score that is harder to game. Van der Zee points to three things lie detection tries to measure: physiological stress, cognitive load, and the conscious efforts people make to conceal lying. Current polygraph technology tackles only one.
But past multi-modal projects have quietly faded. The EU-funded iBorderCtrl program and a US border tool called AVATAR both incorporated video analysis, eye tracking, and body movement detection. Neither is active today.
Kyri Kotsoglou, a legal scholar at Northumbria University, called combining AI and the polygraph "the worst of both worlds" because it adds uncertainty on top of invalidity. Even if machine learning spots previously unseen patterns in physiological data, it cannot reliably link them to lying because there is no real ground truth.
"Even if you have all the records in the world from polygraph tests, you don't know whether those polygraph tests are right or not," said Marion Oswald, a professor of law who has written with Kotsoglou on polygraph use in the justice system. She fears new forms of lie detection will, like the polygraph, be used more as a psychological prop than a scientific tool.
A political context of leak investigations
The push for Polygraph+ comes at a time of high tension within the Pentagon. Under Defense Secretary Pete Hegseth, the department has increasingly turned to polygraph tests to find sources of alleged leaks to the press. In September, the New York Times reported that around 50 officers on the Joint Staff had been given polygraph tests after news coverage about the depletion of US weapons stockpiles in the war with Iran.
Oswald sees the program as a direct response to those pressures. "[Lie detection is] being used as a threat, to intimidate and force people to confess to things, as opposed to anything that's actually getting valid information," she said.
Why this matters for government, science, and research professionals
For those working in federal agencies, national labs, or research institutions that require security clearances, Polygraph+ signals an intent to scale up deception screening with unproven technology. The 2003 NRC report's warning remains relevant: even a system claiming high accuracy will generate large numbers of false positives when applied across a workforce of millions. Researchers and analysts who advise on procurement or ethics reviews should watch whether Congress funds the program and what validation standards DCSA proposes - given that decades of peer-reviewed research have failed to establish a reliable physiological signature of deception. Professionals involved in AI for Government Courses may find this case a useful example of the gap between algorithmic capability and scientific validity in high-stakes deployment.
Your membership also unlocks:
