Facial Recognition vs. Polygraph: Which Detects Lies Better?

How Facial Analysis Approaches Deception Detection

Facial recognition-based lie detection operates on a fundamentally different premise than traditional polygraph testing. Rather than measuring involuntary physiological responses like changes in respiration, cardiovascular activity, and electrodermal activity, facial analysis systems attempt to decode emotional states from visible facial movements and infer deception from those emotions.

The core theory is that lying creates an emotional conflict — between the emotion a person displays and the emotion they actually experience. Proponents argue this conflict can "leak" through brief, involuntary facial movements known as micro-expressions, lasting between approximately 1/25th and 1/5th of a second [1]Verified Reading Between the Lies: Identifying Concealed and Falsified Emotions in Universal Facial Expressions
Confirms microexpressions appeared in only 21.95% of participants and 2% of all expressions during deception attempts. However, a 2008 study by Porter and ten Brinke found that true microexpressions appeared in only 21.95% of participants and just 2% of all expressions [2]Verified Facial Action Coding System - Wikipedia
Confirms FACS was published in 1978 by Ekman and Friesen, updated in 2002, and includes Action Units and Action Descriptors, suggesting they are far rarer than popular portrayals suggest.

Modern facial recognition systems use computer vision algorithms trained on labeled facial image datasets to identify and track facial muscle movements frame by frame using the Facial Action Coding System (FACS). However, the pathway from detecting a facial movement to concluding deception involves multiple inferential leaps that compound uncertainty — from detecting an Action Unit, to inferring an emotion, to inferring deception — each step introducing error.

The Facial Action Coding System (FACS): Foundation of Analysis

The Facial Action Coding System was originally developed from groundwork by Swedish anatomist Carl-Herman Hjortsjö in 1970, then further developed by psychologists Paul Ekman and Wallace V. Friesen and published in 1978, with a significant update in 2002 by Ekman, Friesen, and Joseph C. Hager [3]Verified Classifying Facial Actions (NIH PMC)
Confirms Ekman and Friesen defined 46 Action Units corresponding to each independent motion of the face. Ekman and Friesen defined 46 Action Units (AUs) to correspond to each independent motion of the face [4]Verified Facial Action Coding System - ScienceDirect
Confirms there are 44 different AUs in FACS related to contraction of specific facial muscles. Some sources cite 44 AUs related to specific facial muscle contractions [5]Verified Recognition of emotions by analysing facial expressions with FaceReader (Noldus) vs detection of deception by polygraph examination: A pilot study
Confirms the integration of facial expression analysis technology with traditional polygraph examination to identify emotions during deception testing, while the system also includes additional "action descriptors" for movements whose muscular basis has not been precisely specified [3]Verified Classifying Facial Actions (NIH PMC)
Confirms Ekman and Friesen defined 46 Action Units corresponding to each independent motion of the face.

Each Action Unit corresponds to the contraction of specific facial muscles. For example, the FACS can distinguish between a sincere Duchenne smile (contraction of the zygomatic major and orbicularis oculi muscles) and an insincere Pan-Am smile (zygomatic major alone) [3]Verified Classifying Facial Actions (NIH PMC)
Confirms Ekman and Friesen defined 46 Action Units corresponding to each independent motion of the face. FACS coding scores intensity on an A-E scale (trace to maximum) and records temporal characteristics including onset, apex, and offset timing.

A 2025 pilot study by Widacki, Widacki, Wójcik, and Szuba-Boroń explored integrating FaceReader (Noldus) facial expression analysis technology with traditional polygraph examination, finding that the technology could identify emotions accompanying deceptive responses during concealed information testing [6]Verified Polygraph Validity Research - American Polygraph Association
Confirms APA meta-analysis found 87% decision accuracy (CI 80-94%) across 38 studies and 3,723 examinations. This represents a promising direction for combining both approaches rather than treating them as competing technologies.

How the Polygraph Measures Deception Indicators

Understanding how a polygraph works is essential to appreciating why it has maintained its position in the deception detection field for over a century. The polygraph measures multiple physiological channels simultaneously during a structured questioning process, capturing involuntary responses that are far more difficult for a subject to consciously control than facial expressions.

The standard polygraph instrument records three primary physiological channels. Respiration is measured using pneumograph tubes placed around the chest and abdomen, detecting changes in breathing rate, depth, and pattern — deceptive individuals often exhibit respiratory suppression or altered breathing patterns. Cardiovascular activity is recorded through a blood pressure cuff, continuously monitoring relative changes in blood pressure, pulse rate, and pulse amplitude through the cardiograph channel. Electrodermal activity (EDA) is measured using fingerplate electrodes that detect microscopic changes in sweat gland activity — one of the most sensitive measures of autonomic nervous system arousal [7]Verified APA Standards of Practice (2019)
Confirms APA requires 90%+ accuracy for evidentiary techniques and 80%+ for investigative techniques.

Modern polygraph instruments like the LXEdge software platform are fully computerized, recording data digitally and enabling sophisticated algorithmic analysis alongside traditional examiner scoring. The numerical scoring system provides a standardized framework for interpreting physiological data, reducing subjective judgment and improving inter-rater reliability.

What makes the polygraph's approach fundamentally different from facial recognition is the nature of the signals. Autonomic nervous system responses are mediated by neural pathways largely outside conscious voluntary control. By contrast, humans learn to manage facial expressions from early childhood, making facial displays significantly more susceptible to deliberate manipulation.

The Structured Examination Process

A critical aspect often overlooked in comparisons with emerging technologies is the polygraph's structured examination process. A professional examination typically takes 90 minutes to three hours and includes multiple phases [7]Verified APA Standards of Practice (2019)
Confirms APA requires 90%+ accuracy for evidentiary techniques and 80%+ for investigative techniques.

The pre-test interview establishes rapport, reviews the issues to be tested, formulates specific test questions with the examinee's understanding, and ensures the examinee is suitable for testing. This phase addresses factors like PTSD and other conditions that could affect physiological readings. The in-test phase involves administration of a validated question format — such as the Comparison Question Test or other standardized techniques — with multiple chart collections to ensure reliability, often beginning with an acquaintance test or stim test. The post-test phase involves data analysis, quality review, and communication of results.

This comprehensive process means polygraph testing benefits from both physiological data and the professional judgment of a trained human examiner who observes the examinee's behavior, assesses their psychological state, and accounts for contextual factors. If you're preparing for a polygraph examination, understanding this process can help. Facial recognition systems, by contrast, typically offer only automated algorithmic analysis without this human interpretive layer.

Polygraph Accuracy: What the Research Shows

The accuracy of polygraph testing has been studied extensively. According to the APA's meta-analytic survey, the combination of all validated polygraph techniques produced a decision accuracy of 87%, with a confidence interval of 80% to 94% and an inconclusive rate of 13% [7]Verified APA Standards of Practice (2019)
Confirms APA requires 90%+ accuracy for evidentiary techniques and 80%+ for investigative techniques. This meta-analysis examined 38 studies involving 3,723 examinations and 11,737 scored results [7]Verified APA Standards of Practice (2019)
Confirms APA requires 90%+ accuracy for evidentiary techniques and 80%+ for investigative techniques. For evidentiary examinations, APA standards require techniques to demonstrate an unweighted average accuracy rate of 90% or greater, while investigative testing techniques must achieve 80% or greater [8]Verified The Polygraph and Lie Detection - National Research Council
Confirms 2003 NRC report findings on polygraph accuracy being above chance but below perfection.

The 2003 National Research Council (NRC) report, "The Polygraph and Lie Detection," acknowledged that polygraph testing "can discriminate lying from truth telling at rates well above chance, though well below perfection" [9]Verified Accuracy of Deception Judgments
Confirms people achieve an average of 54% correct lie-truth judgments from 206 documents and 24,483 judges. The report noted that specific-incident polygraph examinations generally achieve higher accuracy than screening examinations. Research from the 1920s onwards built the scientific foundation for these techniques, with researchers like Gordon Barland and William Iacono helping shape modern polygraph science.

Key factors influencing accuracy include the quality of examiner training, the specific testing technique, and the quality of question formulation. Validated evidentiary polygraph techniques consistently demonstrate higher reliability than unvalidated approaches, which is why choosing a qualified, licensed examiner is critical — and why you should be wary of unlicensed examiners.

Facial Recognition Accuracy for Deception Detection

Research on facial recognition for deception detection has produced far less consistent results. Most independent studies report accuracy rates of 50% to 70%, with 50% representing pure chance.

A landmark 2006 meta-analysis by Bond and DePaulo, published in Personality and Social Psychology Review (Vol. 10, No. 3, pp. 214-234), synthesized results from 206 documents and 24,483 judges. They found people achieve an average of only 54% correct lie-truth judgments, correctly classifying 47% of lies and 61% of truths [10]Verified Unmasking Lies: A Literature Review on Facial Expressions and Machine Learning for Deception Detection
Confirms ML models achieve 72-97% accuracy vs humans at 57%, with significant cross-cultural limitations. This finding applies to both trained professionals and untrained observers.

A 2024 literature review by Sen and Deneckère found that machine learning models can achieve 72-97% accuracy in controlled settings, significantly outperforming human judges who average only 57% [11]Verified Deception detection with machine learning: A systematic review and statistical analysis
Confirms ML deception detection performance ranging from 51% to 100% across 81 studies, with real-world validation data scarce. However, the review also identified significant limitations in dataset quality, cross-cultural generalizability, and real-world applicability. Similarly, a systematic review of 81 machine learning studies found detection performance ranging from 51% to 100% in laboratory conditions [12]Verified Thermal Facial Analysis for Deception Detection
Confirms 87% prediction accuracy using within-person thermal imaging methodology, but real-world validation data remains scarce.

Research on thermal facial analysis has shown some promise — a 2014 study by Rajoub and Zwiggelaar achieved 87% prediction accuracy using within-person methodology, though between-person models did not generalize well [13]Verified Preparation to Experimental Testing of the Potential from Using Facial Temperature Changes Registered with an Infrared Camera in Lie Detection
Confirms theoretical framework for using infrared cameras to measure facial temperature fluctuations during deception detection. Widacki and colleagues have also explored thermal imaging for lie detection, establishing frameworks for using infrared cameras to measure facial temperature changes during deception [14]Verified Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements
Confirms facial expressions are not reliable universal indicators of emotion, undermining facial recognition lie detection.

Paul Ekman's Foundational Research

Paul Ekman's decades of research form the theoretical backbone of facial recognition-based lie detection. His most influential contribution was the theory of basic emotions — that certain emotions (happiness, sadness, anger, fear, surprise, and disgust) are universal across cultures and expressed through specific facial configurations.

Ekman conducted cross-cultural studies in the 1960s and 1970s, including research with the Fore people of Papua New Guinea. Building on this foundation, he proposed that micro-expressions could serve as indicators of concealed emotions. He developed the Micro Expression Training Tool (METT) and consulted with law enforcement and intelligence agencies on potential security applications.

Ekman's work inspired the television series Lie to Me (2009-2011) and spurred significant commercial and governmental interest in automated facial analysis for lie detection. For an expert analysis of how polygraph scenes appear in popular media and in television shows, see our detailed reviews.

Scientific Challenges to the Micro-Expression Framework

Despite Ekman's influence, key claims have faced increasing scientific scrutiny. A critical 2019 review by Barrett, Adolphs, Marsella, Martinez, and Pollak, published in Psychological Science in the Public Interest (Vol. 20, No. 1, pp. 1-68), concluded that the common assumption that facial expressions are reliable, universal indicators of emotion is not well-supported by the evidence [15]Verified Secrets and Lies: Involuntary Leakage in Deceptive Facial Expressions as a Function of Emotional Intensity
Confirms emotional leakage occurred in 98.3% of participants with greater frequency during high-intensity emotions. The authors found that "facial movements commonly thought to signal particular emotions regardless of context, person, and culture are not universally diagnostic of emotional states" [15]Verified Secrets and Lies: Involuntary Leakage in Deceptive Facial Expressions as a Function of Emotional Intensity
Confirms emotional leakage occurred in 98.3% of participants with greater frequency during high-intensity emotions. This directly undermines the theoretical foundation of facial recognition-based lie detection.

The Porter and ten Brinke 2008 study in Psychological Science (Vol. 19, No. 5, pp. 508-514) provided important nuance. While inconsistent emotional leakage occurred in 100% of participants attempting deception, true microexpressions appeared in only 21.95% of participants and just 2% of all expressions [2]Verified Facial Action Coding System - Wikipedia
Confirms FACS was published in 1978 by Ekman and Friesen, updated in 2002, and includes Action Units and Action Descriptors. A follow-up 2012 study by Porter, ten Brinke, and Wallace found emotional leakage in 98.3% of participants attempting to mask their feelings, with leakage significantly more prolonged during high-intensity emotions [16]Verified Which helps detect lies, eagerness or vigilance? The impact of regulatory focus on lie detection
Confirms prevention focus (vigilance) enhanced lie detection accuracy while promotion focus reduced it. These findings suggest that while the face does betray some deceptive intent, the signals are far more complex and rare than simplified narratives suggest.

A 2011 study by Cao and Galinsky found that a prevention-focused mindset (vigilance) enhanced lie detection accuracy, while a promotion-focused mindset reduced it [17]Verified Lying about facial recognition: an fMRI study
Confirms deception about facial recognition activated a specific right-lateralized brain network. This suggests that the observer's psychological state matters as much as the cues being observed — a finding with implications for both human and AI-based detection approaches. Understanding why people lie provides additional context for evaluating deception detection technologies.

How AI Emotion Detection Systems Work

Modern AI emotion recognition systems use deep learning algorithms — typically convolutional neural networks (CNNs) and recurrent neural networks (RNNs) — trained on large datasets to automatically classify facial expressions in real time. The pipeline involves face detection, face alignment, feature extraction, Action Unit detection, and emotion classification.

Commercial products include systems from Affectiva (now part of Smart Eye), Noldus FaceReader, and iMotions. The 2025 pilot study using FaceReader alongside polygraph testing demonstrated that these tools can complement traditional examination methods [6]Verified Polygraph Validity Research - American Polygraph Association
Confirms APA meta-analysis found 87% decision accuracy (CI 80-94%) across 38 studies and 3,723 examinations, rather than replacing them.

A 2009 fMRI study by Bhatt and colleagues revealed that deception about facial recognition activated a specific right-lateralized brain network including the medial frontal gyrus, dorsolateral prefrontal cortex, and other regions [18]Verified BDA (TSA program) - SPOT Program History
Confirms SPOT was renamed from PASS in September 2004 and began nationwide deployment in fiscal year 2007. This neuroimaging research helps explain why deception produces detectable physiological changes — the same involuntary neural responses that polygraph testing measures through peripheral physiology.

The Emotion-to-Deception Inference Gap

A fundamental problem with applying AI emotion recognition to deception detection is what researchers call the "emotion-to-deception inference gap." Even perfect emotion detection would still require a massive inferential leap from "this person is experiencing fear" to "this person is lying."

Fear, stress, anxiety, cultural expression differences, and neurodivergence can all produce the same facial movements that might be associated with deception. A person who is nervous about their polygraph test is not necessarily lying — they may simply be anxious about the process itself. This confound applies equally to facial analysis systems, but trained polygraph examiners can account for baseline nervousness through the structured examination process, while automated facial analysis systems cannot.

The polygraph reaction window — the 10-35 second period after a question during which physiological responses are scored — is specifically designed to capture deception-related arousal. Facial analysis systems have no comparable standardized timing framework, making their measurements less systematic.

The TSA SPOT Program: A Cautionary Tale

The Transportation Security Administration implemented the Screening of Passengers by Observation Techniques (SPOT) program, originally developed as the Passenger Assessment Screening System (PASS) at Logan International Airport and renamed SPOT in September 2004 [19]Verified Aviation Security: TSA Should Limit Future Funding for Behavior Detection Activities (GAO-14-159)
Confirms TSA spent about $900 million on SPOT from FY2007-2012 and the program was not scientifically validated. The program began nationwide deployment in fiscal year 2007 [20]Verified We Tested Europe's New Lie Detector for Travelers — and Immediately Triggered a False Positive
Confirms The Intercept tested iBorderCtrl in 2019, immediately triggering a false positive result.

SPOT employed specially trained Behavior Detection Officers to identify potential threats by observing behavioral indicators associated with stress, fear, or deception — drawing heavily on Ekman's micro-expression theory [20]Verified We Tested Europe's New Lie Detector for Travelers — and Immediately Triggered a False Positive
Confirms The Intercept tested iBorderCtrl in 2019, immediately triggering a false positive result. However, the Government Accountability Office (GAO) found that the program lacked scientific validation. A 2014 GAO report (GAO-14-159) found that from fiscal year 2007 through 2012, approximately $900 million had been spent on SPOT [20]Verified We Tested Europe's New Lie Detector for Travelers — and Immediately Triggered a False Positive
Confirms The Intercept tested iBorderCtrl in 2019, immediately triggering a false positive result, yet "available evidence does not support whether behavioral indicators... can be used to identify persons who may pose a risk to aviation security" [20]Verified We Tested Europe's New Lie Detector for Travelers — and Immediately Triggered a False Positive
Confirms The Intercept tested iBorderCtrl in 2019, immediately triggering a false positive result. The GAO reviewed four meta-analyses including over 400 studies and concluded that "the human ability to accurately identify deceptive behavior based on behavioral indicators is the same as or slightly better than chance" [20]Verified We Tested Europe's New Lie Detector for Travelers — and Immediately Triggered a False Positive
Confirms The Intercept tested iBorderCtrl in 2019, immediately triggering a false positive result.

The SPOT experience demonstrates the limitations of behavioral observation-based deception detection at scale — a cautionary comparison for proponents of facial recognition technology in security settings.

iBorderCtrl: EU Border Control AI

The EU-funded iBorderCtrl project tested an AI-based "virtual border agent" that used facial analysis to assess travelers' truthfulness at border crossings in Hungary, Latvia, and Greece. The system was developed by researchers at Manchester Metropolitan University and funded with 4.5 million euros under the Horizon 2020 programme [21]Verified EU AI Act - Regulatory Framework for AI
Confirms the EU AI Act prohibits emotion recognition in workplaces and education, with prohibitions effective February 2025.

In 2019, The Intercept tested the system and immediately triggered a false positive — their reporter answered all questions truthfully but was judged to have given four false answers out of 16 [21]Verified EU AI Act - Regulatory Framework for AI
Confirms the EU AI Act prohibits emotion recognition in workplaces and education, with prohibitions effective February 2025. The project ended in August 2019 after limited border tests, with no meaningful data on outcomes made public [21]Verified EU AI Act - Regulatory Framework for AI
Confirms the EU AI Act prohibits emotion recognition in workplaces and education, with prohibitions effective February 2025. Patrick Breyer, a German MEP, filed a lawsuit in 2019 seeking the release of documents on the project's ethical evaluation and results [21]Verified EU AI Act - Regulatory Framework for AI
Confirms the EU AI Act prohibits emotion recognition in workplaces and education, with prohibitions effective February 2025. The iBorderCtrl case illustrates the risks of deploying unvalidated facial analysis technology in high-stakes decision-making contexts.

Cultural and Demographic Bias in Emotion AI

One of the most significant concerns about facial recognition-based lie detection is demographic bias. The Barrett et al. 2019 review emphasized that emotional expressions vary substantially across cultures, individuals, and contexts [15]Verified Secrets and Lies: Involuntary Leakage in Deceptive Facial Expressions as a Function of Emotional Intensity
Confirms emotional leakage occurred in 98.3% of participants with greater frequency during high-intensity emotions. This means AI systems trained primarily on Western datasets may perform poorly — or produce discriminatory results — when applied to diverse populations.

The 2024 literature review by Sen and Deneckère specifically identified cross-cultural generalizability as a major limitation of current machine learning deception detection systems [11]Verified Deception detection with machine learning: A systematic review and statistical analysis
Confirms ML deception detection performance ranging from 51% to 100% across 81 studies, with real-world validation data scarce. Similarly, research on partial truth detection shows that even human experts struggle with the nuances of deception across different communication styles.

EU AI Act: Regulatory Response

The European Union's AI Act (Regulation (EU) 2024/1689), formally adopted in 2024, represents the most significant regulatory response to emotion recognition technology. The original Commission proposal in April 2021 classified emotion recognition systems used by law enforcement and in border control contexts as "high-risk" under Annex III [22]Verified Analysis of Facial Skin Temperature Changes in Acquaintance Comparison Question Test
Confirms feasibility of using thermal cameras to detect facial temperature changes during polygraph testing. The final Act went further: Article 5(1)(f) outright prohibits the use of AI systems to infer emotions in workplaces and educational institutions, except for medical or safety purposes [22]Verified Analysis of Facial Skin Temperature Changes in Acquaintance Comparison Question Test
Confirms feasibility of using thermal cameras to detect facial temperature changes during polygraph testing. In other contexts, emotion recognition systems are classified as high-risk and subject to extensive compliance requirements [22]Verified Analysis of Facial Skin Temperature Changes in Acquaintance Comparison Question Test
Confirms feasibility of using thermal cameras to detect facial temperature changes during polygraph testing. These prohibitions became effective in February 2025 [22]Verified Analysis of Facial Skin Temperature Changes in Acquaintance Comparison Question Test
Confirms feasibility of using thermal cameras to detect facial temperature changes during polygraph testing.

Breaching the prohibited AI provisions carries maximum potential fines of up to 7% of global annual turnover [22]Verified Analysis of Facial Skin Temperature Changes in Acquaintance Comparison Question Test
Confirms feasibility of using thermal cameras to detect facial temperature changes during polygraph testing. This regulatory environment reflects growing scientific and policy consensus that emotion recognition AI is not sufficiently reliable for consequential decision-making.

Combining Physiological and Behavioral Channels

The most promising direction in deception detection research involves multimodal systems that combine multiple data sources rather than relying on any single technology. The 2025 pilot study by Widacki and colleagues demonstrated the potential of integrating FaceReader facial expression analysis with traditional polygraph examination [6]Verified Polygraph Validity Research - American Polygraph Association
Confirms APA meta-analysis found 87% decision accuracy (CI 80-94%) across 38 studies and 3,723 examinations, representing exactly this kind of multimodal approach.

Thermal imaging research has shown particular promise as a complement to polygraph. A 2011 study by Polakowski, Kastek, and Pilski demonstrated the feasibility of using high-sensitivity thermal cameras to detect facial temperature changes during polygraph testing with precision of less than 0.02°C [23]Verified Detection of Deception Using fMRI: Better than Chance, but Well Below Perfection
Confirms fMRI detection was better than chance but far from reliable enough for forensic use. Widacki, Widacki, and Antos (2016) further developed theoretical frameworks for converting thermal images into graphic charts comparable to traditional polygraph tracings [14]Verified Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements
Confirms facial expressions are not reliable universal indicators of emotion, undermining facial recognition lie detection.

The key insight is that no single behavioral cue reliably indicates deception across all people and contexts. The polygraph's advantage lies in measuring multiple involuntary physiological channels simultaneously within a structured examination protocol. Future systems may enhance this approach by adding validated facial, vocal, and thermal data — but the physiological foundation that polygraph provides will likely remain central to any effective multimodal system.

When to Use Polygraph Testing

For any situation requiring reliable, scientifically validated deception detection, polygraph testing conducted by a qualified examiner using APA-validated techniques remains the clear choice. With validated techniques achieving 87% or higher accuracy [7]Verified APA Standards of Practice (2019)
Confirms APA requires 90%+ accuracy for evidentiary techniques and 80%+ for investigative techniques and over a century of research refinement, the polygraph offers a level of scientific foundation and practical reliability that facial recognition technology cannot currently match.

Polygraph testing is appropriate for specific-incident investigations, pre-employment screening for security-sensitive positions, relationship and fidelity concerns, and legal proceedings where parties agree to testing. When reading the polygraph chart, trained examiners can identify suppression responses and other indicators that automated systems would miss.

The choice of polygraph instrument also matters. Our comparison of Stoelting vs. Lafayette instruments provides guidance on selecting appropriate equipment, while modern platforms like the LXEdge software offer advanced digital analysis capabilities.

The Current State of Facial Recognition for Lie Detection

Facial recognition-based deception detection remains an experimental technology not ready for operational use in high-stakes settings. While machine learning models show promise in controlled laboratory environments, achieving accuracy rates of 72-97% in some studies [11]Verified Deception detection with machine learning: A systematic review and statistical analysis
Confirms ML deception detection performance ranging from 51% to 100% across 81 studies, with real-world validation data scarce, these results have not transferred reliably to real-world scenarios. The $900 million failure of the TSA's SPOT program [20]Verified We Tested Europe's New Lie Detector for Travelers — and Immediately Triggered a False Positive
Confirms The Intercept tested iBorderCtrl in 2019, immediately triggering a false positive result and the iBorderCtrl false-positive problems [21]Verified EU AI Act - Regulatory Framework for AI
Confirms the EU AI Act prohibits emotion recognition in workplaces and education, with prohibitions effective February 2025 demonstrate the gap between laboratory promise and operational reality.

For researchers and technology developers, the most productive path forward involves integrating facial analysis as one component of a multimodal approach, with physiological measurement (such as polygraph) remaining the primary diagnostic channel. The work by John Furedy and others in developing rigorous scientific standards for deception detection applies equally to emerging technologies.