# Pentagon Seeks $30 Million for AI-Enhanced Lie Detector, and Experts Are Skeptical

A US defence budget request seeks $30.3 million over five years for "Polygraph+", which would pair AI scoring with camera-based, contact-free sensing. Researchers warn the science behind lie detection remains weak.

## A new push to modernise the polygraph
A defence budget request has revealed plans to spend **$30.3 million over five years** on an upgraded form of lie detector. The programme is called **Polygraph+** (also referred to as Polygraph Next) and is managed by the US Defense Counterintelligence and Security Agency. Its stated purpose is to help vet employees and detect insider threats, meaning people with trusted access who might misuse it.

The effort is reported by MIT Technology Review. It is a useful case study in a wider trend: organisations are increasingly turning to artificial intelligence to assess people, not just systems, and that raises hard questions about accuracy and fairness.

## What the programme would change
Traditional polygraph tests record physiological signals such as heart rate, breathing and skin conductivity while a person answers questions. An examiner then interprets the results. Polygraph+ is built around two main ideas:

- **AI and machine-learning scoring algorithms** that would analyse the recorded signals and produce a score.
- **"Standoff sensing"**, which means collecting physiological readings without attaching sensors to the person.
According to the report, two companies have built prototypes. Presage Technologies measures heart rate and breathing using ordinary cameras. Altec Research looks at head movement, facial skin temperature and pore activity.

## Why researchers are doubtful
The main objection is not about the technology being new, but about the underlying premise. Polygraphs measure stress-related bodily responses, and those responses can have many causes besides lying. Nervousness, fear of being wrongly accused, or simply an intimidating setting can all change a person's heart rate and breathing.

The report cites a 2003 review by the US National Research Council that found the evidence for polygraph effectiveness to be "weak at best". It also notes research showing that people, unaided, detect lies only slightly more than half the time.

Several academics quoted in the article are critical of the plan:

- Kyri Kotsoglou described it as "a misguided effort to reduce the complex to something that is tangible".
- Sophie van der Zee observed that "there is still no Pinocchio's nose", meaning no single reliable physical signal of deception has been found.
- Marion Oswald warned that combining AI with polygraphs risks creating "the worst of both worlds".

## Context: more polygraph use
The budget request follows an increase in polygraph testing within the Pentagon under Defense Secretary Pete Hegseth. The article mentions that around 50 Joint Staff officers were tested in connection with alleged leaks about weapons stockpiles.

## Why this matters beyond one department
Even if you never take a polygraph, the story touches on issues that affect many people:

- **Automation can add false confidence.** A numerical score from an algorithm can look objective even when the signal it analyses is weak. If a method is unreliable, adding machine learning does not necessarily fix it, and it can make errors harder to question.
- **Contact-free sensing changes consent.** A conventional test requires a person to sit down and be connected to equipment. Camera-based methods could, in principle, make it easier to collect physiological data with less obvious notice. How such data would be stored, shared and protected is an open question.
- **Consequences can be serious.** Screening results can influence careers, security clearances and investigations. False positives and false negatives both carry real costs.
- **Workplace tools tend to spread.** Techniques developed for high-security settings sometimes migrate to ordinary employers, border checks or hiring tools.

## What readers can do
- **Ask for evidence.** When a product claims to detect lies, emotions or intent using AI, ask for independent, peer-reviewed validation and published error rates.
- **Look for human oversight.** Automated scores should be one input among many, with a clear way to challenge results.
- **Understand data rights.** If your employer or another organisation uses biometric or physiological monitoring, find out what is collected, how long it is kept and who can see it.
- **Support clear rules.** Organisations adopting AI-based assessment tools should have governance policies covering accuracy testing, consent, bias and appeal processes.

## The bottom line
Polygraph+ is still a budget proposal and prototype effort, and the article does not report any proven results. What it does show is a familiar pattern: a long-contested technique is being given a modern AI makeover, while scientists caution that the core scientific problem, reliably telling truth from lies through body signals, remains unsolved. Readers should treat claims of AI-powered "lie detection" with healthy scepticism until robust evidence supports them.

Source: [MIT Technology Review](https://www.technologyreview.com/2026/09/25/1145144/pentagon-ai-lie-detector/)