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

The U.S. Department of Defense (DoD) has requested $30.3 million over five years to develop a new AI-powered lie detector called Polygraph+ or Polygraph Next. This program aims to modernize federal polygraph technology by using artificial intelligence (AI) and machine learning to analyze physiological responses. The project also explores standoff sensing, which allows examiners to measure a person’s heart rate, breathing, or sweat without attaching devices to their body. The goal is to improve the accuracy and reliability of lie detection for vetting federal employees and detecting insider threats. The budget request has not yet been approved by Congress. The Pentagon, under Defense Secretary Pete Hegseth, has increased polygraph use recently, particularly after reports of leaks to the press, including a case where around 50 Joint Staff officers were tested in September 2023.

Current polygraph tests measure physiological responses like blood pressure, pulse, breathing, and sweat to assess truthfulness. Examiners compare a person’s reactions to baseline questions (e.g., "Is the sky blue?") with target questions (e.g., "Have you ever committed a crime?"). The assumption is that lying causes stress, altering these measurements. However, polygraphs have been criticized for their unreliability. The U.S. government conducts tens of thousands of polygraph tests annually, but their results are rarely accepted in court. In 1983, Congress’s Office of Technology Assessment found limited evidence supporting polygraphs for employee screening, and in 2003, the U.S. National Research Council (NRC) called their efficacy "weak at best." The American Polygraph Association claims 80% to 94% accuracy, but the NRC warned that even this rate could lead to many false accusations in large-scale screenings, such as the DoD’s 2.8 million employees.

Polygraphs face several challenges. First, interpretations are subjective: different examiners often reach conflicting conclusions. Second, people from minority groups are more likely to be judged as deceptive. Third, individuals can learn countermeasures to manipulate results, such as stepping on a hidden pin to alter their physiological responses. Sophie van der Zee, an associate professor studying deception at Erasmus University, notes that the biggest effect of polygraphs is often deterrence—people confess before the test even begins. However, this only works if they believe the polygraph is reliable. Over decades, various technologies like thermal cameras, pupil trackers, and brain scans have been tested for lie detection, but none have proven reliable outside lab settings. As van der Zee says, "There is still no Pinocchio’s nose," meaning no single, foolproof sign of lying exists.

AI could theoretically improve lie detection by identifying patterns in physiological data that humans might miss. It may also enable multi-modal deception detection, combining multiple measurements (e.g., heart rate, facial temperature, and head movement) into a single "deception score" to reduce the risk of manipulation. For example, Presage Technologies claims to measure heart and breathing rates using standard cameras, while Altec Research tracks head movement, facial skin temperature, and pore activity. Past projects like the UK’s Silent Talker and the EU’s iBorderCtrl used similar approaches but were discontinued. However, experts like Kyri Kotsoglou and Marion Oswald argue that AI-polygraph hybrids risk combining "the worst of both worlds": added uncertainty atop an already unreliable method. Even with vast data, there’s no way to verify if polygraph results are correct, as there’s no ground truth for deception.