Dutch researchers tested AI lie detection for EU border checks despite scientific doubts
Dutch researchers and a European Union-funded consortium tested AI-based systems for detecting deception at border controls, even as scientists and EU lawmakers warned that lie detection has no reliable scientific basis, Trouw reported.
Between 2018 and 2021, the Dutch applied research institute TNO reportedly worked with Amsterdam-based tech company Vicarvision on a European Commission–funded project exploring whether artificial intelligence could identify when travelers lie during border checks. The system was designed to make border screening more efficient and was intended for use with “high-risk travelers."
The project, later revealed through investigative reporting by Investico for Trouw, De Groene Amsterdammer, and Argos, was part of a broader EU effort to use AI and risk profiling in border management. It reportedly included participation from 25 organizations, among them the Royal Netherlands Marechaussee and the Greek Coast Guard, and received nearly 8 million euros in total funding from the European Commission.
Under the EU AI Act, AI-based “lie detection” is classified as a high-risk application due to potential human rights violations and is generally prohibited for European citizens, with limited exceptions left for criminal investigations or when looking into migrants crossing into Europe. Critics, however, say the legal carve-outs still leave room for controversial deployments.
The project focused on analyzing facial expressions and behavior in controlled settings, including participants playing a card game in which some were instructed to bluff. Researchers examined behaviors such as eye movements, leaning forward or backward, and forced smiling as potential indicators of deception. These findings contributed to the development of a system called the Multi-Modal Communication Analysis Tool, or MMCAT. The effectiveness results of the system remain classified.
TNO and Vicarvision both reject the characterization of the system as a lie detector. A TNO spokesperson said it “detects no lies” but instead identifies “physical behaviors” that could help border officials distinguish “truths, half-truths, and lies.” Vicarvision said the tool was intended only as “support” for interviewing border officers. The European Commission also denies that the project produced a lie detector.
However, Bruno Verschuere, professor of forensic psychology at the University of Amsterdam, said the distinction is misleading. “A lie detector is actually a psychological test that tries, based on a signal, a technique or a measurement, to distinguish between truth and lying,” he said, adding that the work clearly fits that definition.
Verschuere criticized the continued reliance on nonverbal indicators. “We know from hundreds of studies that the link is very weak,” he said. “This time it is with AI, but in essence they keep doing the same thing in a different jacket.”
Vicarvision CEO Tim den Uyl said the company ultimately decided not to pursue lie-detection applications and changed its terms of service to prohibit such use. “I am very happy that we do not make software for border systems and sell it everywhere,” he said. “I would have found it very distressing if, for example, U.S. Immigration and Customs Enforcement (ICE) were using our technology to wrongly deport people.”
He added that control over how research outputs are used is limited once shared. “The technology is ours,” Den Uyl said, “but I have no insight into what others do with the findings.”
The European Commission continues to fund research into AI-assisted border security. A second program, known as TRESSPASS, also involved deception-related research and reflects ongoing EU investment in automated border risk assessment systems.
