Pentagon Seeks $30.3 Million for an AI-Enhanced Lie Detector
The Pentagon is seeking $30.3 million for an AI-enhanced polygraph system, despite longstanding doubts about whether physiological signals can reliably reveal deception.
The US Department of Defense is seeking $30.3 million over five years to develop a new generation of lie-detection technology. Known as Polygraph+ or Polygraph Next, the proposed program would combine artificial intelligence with physiological monitoring, including methods that do not require sensors to be attached to a person.
The project is intended to modernize federal polygraph and credibility-assessment systems. However, it would build on a technology whose scientific reliability has been questioned for decades. Researchers and legal scholars warn that AI cannot resolve the central problem: No physiological response consistently proves that a person is lying.
What the Pentagon is proposing
According to a Department of Defense budget request, Polygraph+ would focus on AI- and machine-learning-based scoring algorithms. It would also explore “standoff sensing,” a term for collecting physiological signals without physically connecting equipment to the person being examined.
The Defense Counterintelligence and Security Agency would manage the program. The agency conducts background checks for the federal government, and the budget document says the technology could be used for:
- Vetting prospective employees
- Detecting potential insider threats
- Improving the accuracy and reliability of credibility assessments
The funding request has not been approved by Congress. The document also does not identify the specific technologies that would ultimately be included in the system.
The proposal comes as the Pentagon has increased its use of polygraph examinations in investigations into alleged press leaks. In September, around 50 officers on the Joint Staff reportedly underwent polygraph testing following coverage about the depletion of US weapons stockpiles during the war with Iran.
Non-contact sensors offer clues to the approach
Earlier Pentagon work indicates what a future system might measure. In 2023, the Defense Innovation Unit invited companies to submit technologies for detecting deception and selected two businesses to develop prototypes.
Presage Technologies says it can use standard cameras to measure heart rate and breathing rate. Altec Research, a medical sensor company expanding into non-contact sensing, developed a prototype that tracks head movement, facial skin temperature and pore activity.
These measurements would broaden the range of signals available to examiners. Traditional polygraphs generally record blood pressure, pulse, breathing and perspiration. Examiners compare responses to baseline questions, such as whether the sky is blue, with reactions to questions about the matter under investigation.
AI could potentially search this larger collection of data for patterns that a human examiner might miss. It could also combine several measurements into a single deception score, an approach known as multimodal detection.
Polygraphs have a disputed scientific record
The basic design of the polygraph has changed little since its invention in the 1920s. Despite continued federal use, its accuracy has repeatedly been challenged, and polygraph results are rarely admissible in court.
In 1983, Congress’s Office of Technology Assessment found very limited evidence supporting polygraphs for employee screening. Two decades later, the US National Research Council described the evidence for their effectiveness as “weak at best.”
The American Polygraph Association claims accuracy of between 80% and 94%. However, the National Research Council warned that even a screening system performing at that level could produce many errors when applied to a large workforce. The Defense Department employs 2.8 million people, meaning an imperfect test could falsely identify tens of thousands at that scale.
Polygraph testing also has practical weaknesses:
- Interpretations can vary significantly between examiners.
- People from minority groups are more likely to be assessed as deceptive.
- Trained subjects can use countermeasures to manipulate their physiological responses.
- Stress or fear may produce reactions that resemble supposed signs of deception.
One countermeasure is to deliberately increase the response to baseline questions, such as by stepping on a concealed pin. This can make later comparisons less useful.
Why adding AI may not solve the core problem
Researchers have experimented with thermal cameras, pupil tracking, brain scans, eye tracking, voice analysis and body-movement monitoring. These approaches have not produced consistently reliable results outside laboratory settings.
The underlying challenge is not simply a lack of data. Deception has no universal physical marker. Physiological stress, cognitive effort and a person’s conscious attempt to hide a lie may all affect behavior, but none provides a definitive signal in every person or situation.
Sophie van der Zee, an associate professor at Erasmus University in Rotterdam who studies deception, argues that combining methods aimed at these different processes could improve performance. Similar ideas have been tested before. Silent Talker, developed by researchers at Manchester Metropolitan University, generated deception scores from video. It later became part of the EU-funded iBorderCtrl pilot. In the US, AVATAR combined eye tracking, voice analysis and body-movement detection for use at border crossings. Those projects subsequently faded from use.
Legal scholar Kyri Kotsoglou of Northumbria University describes combining AI with polygraphs as adding uncertainty to an already invalid foundation. Even if an algorithm discovers new patterns, researchers still need reliable labels showing which subjects were actually lying. Existing polygraph records cannot provide that ground truth if the original judgments may have been wrong.
A tool for evidence or psychological pressure?
Critics also question whether the main effect of lie-detection systems is scientific or psychological. Van der Zee says polygraphs can deter people or prompt confessions before an examination begins, but that effect depends on subjects believing the technology works.
Marion Oswald, a law professor who has written with Kotsoglou about polygraphs in the justice system, warns that newer systems could similarly function as psychological props. In the Pentagon’s current environment, she argues, lie detection may be used to pressure or intimidate people over leaks and perceived disloyalty rather than to obtain valid information.
Polygraph+ could make credibility assessments more automated and less physically intrusive. Yet non-contact sensors and advanced algorithms do not answer the essential scientific question of whether measured behavior reliably indicates deception. Until that link can be demonstrated, the proposed system will carry forward many of the limitations associated with traditional polygraphs.
Original reporting: revew
Originally reported by revew.