The Pentagon requests $30 million to build an AI-powered lie detector

The United States Department of Defense has requested $30.3 million over five years to develop an improved version of the lie detector, according to a budget request first reported by the outlet Inside Defense. The program, dubbed Polygraph+ or Polygraph Next, will focus on two fronts: scoring algorithms based on artificial intelligence and machine learning, and a technique called 'standoff sensing', which makes it possible to take physiological readings without connecting any device to the body of the person being examined. According to the budget document, the stated goal is to 'modernize federal polygraph and credibility assessment technologies' to improve their accuracy and reliability.

The project comes at a time of intense internal tension at the Pentagon. Under Defense Secretary Pete Hegseth, the department has increasingly turned to polygraph tests to try to identify the source of leaks to the press. In September, the New York Times reported that around 50 officers of the Joint Chiefs of Staff were subjected to polygraph tests after information was published about the depletion of US weapons stockpiles during the war with Iran.

Polygraph+ will be managed by the Defense Counterintelligence and Security Agency (DCSA), which is responsible for carrying out background checks for the federal government. According to the document, which has not yet been approved by Congress, the new technology would be used both for screening prospective employees and for 'insider threat detection'. It is still unclear which specific technologies will be used, and the DCSA did not respond to a request for further information.

Other previous Pentagon projects offer clues about the kind of tools that could be incorporated. In 2023, the Defense Innovation Unit (DIU) launched an open solicitation process to find companies with products applicable to deception detection, and selected two to build prototypes: Presage Technologies, which claims it can measure heart rate and respiratory rate using conventional cameras, and Altec Research, a medical sensor company now moving into contactless detection technologies. A screenshot of Altec's prototype released by the DIU shows that the tool tracks head movement, facial skin temperature and pore activity. Neither Presage Technologies nor Altec Research responded to requests for comment, and the DIU declined to comment.

Current lie detection technology has barely changed since the polygraph was invented in the 1920s. Examiners rely on measurements of blood pressure, pulse, breathing and sweating, comparing the physiological response to baseline questions (such as 'is the sky blue?') with target questions (such as 'have you ever committed a crime?'). The federal government conducts tens of thousands of these tests every year for personnel screening, but their reliability has been repeatedly questioned and their results are rarely admissible in court. In 1983, Congress's Office of Technology Assessment concluded that there was very little evidence supporting the use of the polygraph for employee screening, and in 2003 the United States National Research Council (NRC) described the evidence of its effectiveness as 'weak at best'. Various studies suggest that human beings, without technical assistance, are able to detect a lie just over half the time.

The American Polygraph Association maintains that the test is between 80% and 94% accurate. However, the NRC's 2003 report warned that even that level of accuracy, applied at scale, could generate numerous errors: the Department of Defense employs 2.8 million people, so an imperfect system applied at that scale could end up falsely accusing tens of thousands of people. Added to this are other known problems: polygraph interpretations are often subjective, different examiners obtain very different results, and people from minority groups are more likely to be judged deceptive. Moreover, with some training it is possible to learn countermeasures to beat the test; for example, artificially intensifying the physiological response to baseline questions by stepping on a tack hidden in one's shoe.

Sophie van der Zee, an associate professor who studies deception at Erasmus University Rotterdam, notes that 'if you know how it works, you can beat it'. As she explains, the machine's main effect is deterrent: subjects often confess even before the test begins, although that 'only works if people believe the polygraph works'. Over the decades, other lie detection strategies have been tried, from thermal cameras to pupil trackers and brain scanners, without any producing reliable results outside the laboratory. In van der Zee's words, 'the Pinocchio nose does not yet exist'.

Artificial intelligence could, in theory, improve polygraphs if it were able to find patterns in the data that human examiners do not detect. According to van der Zee, there are three underlying phenomena that lie detection tries to identify: physiological stress, cognitive load and conscious efforts to hide that one is lying. Current polygraph technology addresses only one of them, so combining several methods that tackle these three different angles would, in theory, increase the chances of success. This idea is not new: in the 2000s, researchers at Manchester Metropolitan University (United Kingdom) developed a system called Silent Talker, which generated a deception score from video recordings, later incorporated into the EU-funded iBorderCtrl project. In the United States, a project called AVATAR combined eye tracking, voice analysis and body movement detection for use at border crossings. All of these projects have quietly faded away over time.

The experts consulted are skeptical about the Pentagon's new initiative. Kyri Kotsoglou, a law specialist at Northumbria University (United Kingdom) who studies the use of polygraphs in the justice system, describes the project as 'a misguided effort to reduce the complex to something tangible', and warns that combining artificial intelligence with the polygraph represents 'the worst of both worlds', because it adds uncertainty to a technique that already lacks validity. Marion Oswald, a law professor who has written with Kotsoglou about the use of polygraphs in the justice system, stresses that even if every polygraph test record in the world were available, there would be no way to know whether those tests were correct, since there is no ground truth against which to check them.

Oswald fears that these new forms of lie detection will be used, like the polygraph itself, more as a psychological pressure tactic than as a scientific tool. In her view, the initiative 'seems to respond above all to the current administration's concern about leaks and alleged disloyalty', and she warns that lie detection 'is being used as a threat, to intimidate and force confessions, rather than as a means of obtaining genuinely valid information'.