Pentagon’s $30.3 Million AI Lie Detector Plan Tests Polygraph Limits

A $30.3 million Pentagon budget request would fund a five-year effort to modernize lie detection with artificial intelligence, MIT Technology Review reported, putting a contested screening tool at the center of a new federal technology program.

The project is called Polygraph+ or Polygraph Next.

Its focus is a mix of AI and machine-learning scoring algorithms with “standoff sensing,” a method for taking physiological readings without attaching equipment to the person being assessed.

The budget document says the work would modernize federal polygraph and credibility-assessment technologies to improve accuracy and reliability.

The program has not yet been approved by Congress.

The Defense Counterintelligence and Security Agency would run it, and the planned uses include vetting prospective employees and detecting insider threats.

The record does not identify the exact sensors or algorithms the agency would buy, and DCSA did not provide more detail.

The timing gives the proposal a sharper operational context.

Under Defense Secretary Pete Hegseth, the Pentagon has turned to polygraph tests as it tries to identify alleged leakers.

In September, the New York Times reported that around 50 Joint Staff officers had been tested after coverage of depleted U.S. weapons stockpiles during the war with Iran.

Earlier Pentagon work shows the kind of technology that could shape the program.

In 2023, the Defense Innovation Unit sought deception-detection products and selected Presage Technologies and Altec Research to build prototypes.

Presage has claimed standard cameras can measure heart rate and breathing.

Altec’s prototype materials showed tracking for head movement, facial skin temperature and pore activity.

Neither company commented for the source article, and DIU also declined to comment.

That sensor path extends a century-old model rather than replacing the underlying premise.

Traditional polygraph examiners compare blood pressure, pulse, breathing and sweat responses to baseline and target questions.

The federal government uses tens of thousands of these tests each year to screen employees, but the results are rarely admissible in court and have faced repeated scientific challenges.

Congress’s Office of Technology Assessment concluded in 1983 that evidence supporting employee-screening use was very limited.

The National Research Council said in 2003 that evidence for polygraph efficacy was “weak at best.” The scale problem is central: the Defense Department employs 2.8 million people, so a system that looks accurate in percentage terms could still falsely accuse tens of thousands if used broadly for screening.

Supporters of a newer system could argue that AI might find patterns examiners miss, especially when multiple signals are combined into a deception score.

Sophie van der Zee, a deception researcher at Erasmus University in Rotterdam, told MIT Technology Review that lie detection tries to capture physiological stress, cognitive load and conscious concealment, while today’s polygraph addresses only one of those areas.

The history of alternative systems gives the Pentagon less certainty.

Thermal cameras, pupil tracking, brain scans, video scoring systems and border-screening tools have all been tried without producing reliable results outside laboratory conditions.

Van der Zee’s summary was blunt: “There is still no Pinocchio’s nose.”

Legal scholar Kyri Kotsoglou of Northumbria University called the project a misguided attempt to make a complex judgment tangible.

He argued that pairing AI with polygraph methods adds uncertainty to a technique whose validity is already disputed.

Marion Oswald, a law professor who has written with Kotsoglou on polygraphs, warned that new lie-detection tools could become a psychological prop rather than a scientific test.

That is the policy question beneath the procurement line.

If Congress funds Polygraph+, the Pentagon will be buying into a sensor-and-algorithm pathway before the source evidence shows a reliable way to connect physiological patterns with deception.

Without that ground truth, the program’s operational value depends less on whether the technology can produce a score than on whether federal agencies can justify using that score against employees and applicants.