Pentagon Seeks $31 Million for Polygraph+, but Some Voice and Face AI Dropped Out of the Plans

According to a budget request that Congress has not yet approved, the US Department of Defense wants to spend about $31 million (MIT Technology Review reports $30.3 million over five years) on Polygraph+, a program meant to equip the classic lie detector with artificial intelligence algorithms and sensors that work without touching the body. The project is intended for vetting job candidates and detecting insider threats. After reporters asked questions, the agency running the program said that some of the AI tools discussed earlier are not part of the current work.

What the budget covers

According to budget materials described by Defense News, the project includes scoring algorithms, decision aids, thermal imaging that detects stress-related changes in blood flow, and other non-contact sensors. MIT Technology Review, citing the DoD budget request, writes about $30.3 million over five years and a goal of "modernizing federal polygraph and credibility assessment technologies".

The intended uses are vetting job candidates and detecting insider threats. The specific technologies have not been disclosed, and DCSA did not respond to MIT Technology Review's request for comment. Inside Defense was the first to report on the program.

What was left out

In a statement sent to Defense News, DCSA said that active research and development does not include generative AI or large language models. It also excluded micro-expression analysis, facial action coding and vocal analysis, though the department will "continue to monitor emerging scientific capabilities". The agency added that the Air Force is not involved at this stage, nor is the university that drafted the May 2025 chart.

Defense News notes that DCSA did not address other models described on the slides, which scan speech and written statements for "deceptive" patterns. The agency also declined to explain the scope of the project or the reasons for dropping voice and face analysis. Slides from 2021-2025 show that the Air Force, DCSA and academic labs have been working on sentiment analysis and deceptive speech models since 2019.

Training data and promises

An anonymous Pentagon official said in July that AI is meant to support investigators, not replace them, and acknowledged that the models are primarily trained on data from laboratory studies with volunteer participants. An early Air Force model was trained on text from Twitter, Reddit and WordNet. Farakh Zaman of the Air Force Office of Scientific Research said the algorithms may help establish "a more objective baseline", but need further testing and validation before fielding.

A May 2025 slide shows an avatar interviewer: a camera and microphone feed data into language models that gauge the applicant's emotional state, while a human observes the interview and keys in the questions the avatar asks. According to the department, the results are meant only as auxiliary data for certified examiners.

Scientists and lawyers push back

It’s a misguided effort to reduce the complex to something that is tangible. - Kyri Kotsoglou, Northumbria University, on plans to combine AI with the polygraph (MIT Technology Review)

Kotsoglou calls AI lie detection "the worst of both worlds" because it adds uncertainty on top of a method of doubtful validity. Marion Oswald, a professor of law, argues that without knowing the ground truth, even vast archives of polygraph results cannot show whether the tests were right. Sophie van der Zee of Erasmus University Rotterdam reminds us: "There is still no Pinocchio's nose".

David Markowitz of Michigan State University said Google Gemini performed no better than chance in his tests and falsely accused people more often than humans did. Attorney Mark Zaid pointed to due process problems, since the fallout from a failed test can block a career for a year or longer. Critics suggest AI analysis may be better suited to hunting disloyalty than detecting lies. Jay Stanley of the ACLU noted that even if it were purely an intimidation tool, it might not be utterly useless from the point of view of large bureaucracies. Defense News stresses that the European Union classifies such systems as high risk.

The political backdrop

MIT Technology Review notes that under Defense Secretary Pete Hegseth, the Pentagon has turned to the polygraph more often in hunting for leaks. In September, the New York Times reported that around 50 officers on the Joint Staff were tested after news coverage of depleted weapons stockpiles in the war with Iran. For Polish readers, this signals growing pressure in the US for automated assessment of people's credibility, while in the EU similar systems fall under stricter rules of the AI Act.