High-Risk AI Systems—Lie Detection Application

Integrating artificial intelligence into border control systems may help to strengthen security and make operations more efficient. For example, the emerging application of artificial intelligence for lie detection when inspecting passengers presents significant opportunities for future implementation. However, as it makes use of technology that is associated with artificial intelligence, the system is classified as high risk, in accordance with the EU AI Act and, therefore, must adhere to rigorous regulatory requirements to mitigate potential risks. This manuscript distinctly amalgamates the technical, ethical, and legal aspects, thereby offering an extensive examination of the AI-based lie detection systems utilized in border security. This academic paper is uniquely set apart from others because it undertakes a thorough investigation into the categorization of these emerging technologies in terms of the regulatory framework established by the EU AI Act, which classifies them as high risk. It further makes an assessment of practical case studies, including notable examples such as iBorderCtrl and AVATAR. This in-depth analysis seeks to emphasize not only the enormous challenges ahead for practitioners but also the progress made in this emerging field of study. Furthermore, it seeks to investigate threats, vulnerabilities, and privacy concerns associated with AI, while providing security controls to address difficulties related to lie detection. Finally, we propose a framework that encompasses the EU AI Act’s principles and serves as a foundation for future approaches and research projects. By analyzing current methodologies and considering future directions, the paper aims to provide a comprehensive understanding of the viability and consequences of deploying AI lie detection capabilities in border control.

Comprehensive policy and legal analysis examining AI lie detection technologies through the EU AI Act's risk-based framework, evaluating case studies of iBorderCtrl and AVATAR systems, and proposing a three-layered compliance framework with security controls.

Background & Context

The deployment of artificial intelligence in border security represents a critical convergence of technological innovation and fundamental rights protection. AI-based lie detection systems for inspecting passengers at borders are classified as high-risk under the EU AI Act, requiring rigorous regulatory compliance to mitigate potential risks. This 2025 analysis by Kalodanis and colleagues arrives at a pivotal moment as the EU AI Act's prohibitions and high-risk system requirements enter force.

The paper fills a crucial gap by examining how emerging border control technologies fit within the EU's groundbreaking regulatory framework. The study distinctly amalgamates technical, ethical, and legal aspects, investigating how these technologies are categorized as high-risk under the EU AI Act. With systems like iBorderCtrl and AVATAR already piloted at European borders, understanding their regulatory status and associated risks has immediate practical importance for policymakers, border agencies, and AI developers.

Ongoing debates emphasize the need for transparent, accountable, and rights-preserving approaches when integrating AI and biometric technologies into border management, particularly when solutions have significant influence on privacy and human rights. This work bridges the gap between AI capability claims and regulatory reality.

Research Design & Methodology

This is a comprehensive policy and legal analysis rather than an empirical study. The authors employed a multifaceted analytical approach combining regulatory analysis, technical evaluation, and case study review. The methodology centered on examining AI-based lie detection systems through the lens of the EU AI Act's risk-based classification framework.

The research analyzed two major case studies: iBorderCtrl and AVATAR, notable examples of AI lie detection applications in border security. For iBorderCtrl, the authors examined a system that profiles travelers based on computer-automated interviews and AI-based analysis of 38 microgestures, with claimed accuracy of approximately 75% during controlled trials. The AVATAR system analysis covered technology with reported deception-detection success rates of 60 to 75 percent and sometimes up to 80 percent.

Key methodological components included:

• Mapping AI lie detection systems to the EU AI Act's four-tiered risk categorization framework
• Analyzing threats, vulnerabilities, and privacy concerns specific to these technologies
• Developing security controls tailored to high-risk border control applications
• Proposing a three-layered compliance framework incorporating EU AI Act principles

Results & Key Findings

AI lie detection systems for border control qualify as high-risk under the EU AI Act, triggering stringent requirements for technical documentation, conformity assessment, human oversight, and transparency. The analysis revealed significant performance and ethical concerns with existing pilot systems.

For the iBorderCtrl project specifically:

• The system claimed an accuracy rate of approximately 75% during controlled trials, but had a high false-positive rate, incorrectly flagging innocent travelers as deceptive in about 24.45% of cases
• The EUR 4.5 million project analyzed 38 microgestures through webcam interviews to detect illegal immigrants and prevent crime and terrorism
• Testing revealed real-world failures: a journalist providing honest responses was deemed a liar by the machine, with four false answers out of 16 and a score of 48
• The training dataset comprised only 32 individuals (10 Asian/Arabic and 22 White Europeans), raising concerns about bias and generalizability

Studies suggest that without significant improvements in accuracy and ethical safeguards, these systems may not be suitable for deployment, highlighting the need for transparent, accountable approaches. The iBorderCtrl project epitomizes the complexity of deploying advanced technologies in sensitive environments where AI solutions significantly influence privacy and human rights.

Discussion & Significance

This analysis makes a unique contribution by operationalizing the EU AI Act's requirements for a specific high-stakes application. The paper demonstrates that theoretical regulatory frameworks have profound practical implications for border security technology development. The high-risk classification means these systems cannot be deployed without meeting strict technical, ethical, and governance standards that many current systems fail to satisfy.

The work illuminates a fundamental tension: border agencies seek efficiency gains through automation, yet the technologies proposed lack the scientific validity and accuracy needed for high-stakes decisions affecting fundamental rights. Expert assessment deemed the iBorderCtrl project "not credible" because there is no evidence that monitoring microgestures is an accurate way to measure lying. This credibility gap becomes legally consequential under the EU AI Act's requirements.

The authors' proposed three-layered framework addresses compliance requirements, security controls, and ethical governance—providing a roadmap for developers seeking to build legally compliant systems. This matters because the framework lays out cybersecurity and privacy requirements for AI systems deemed high risk, with organizations like OpenAI and Google formally obligated to comply. The paper effectively translates abstract regulatory language into concrete technical and operational requirements for border control applications.

Limitations & Considerations

As a legal and policy analysis rather than empirical research, the paper does not generate new experimental data on lie detection accuracy. The analysis relies on publicly available information about iBorderCtrl and AVATAR, which may be incomplete due to proprietary restrictions and the politically sensitive nature of border security technologies. Internal documents obtained through freedom of information requests were heavily redacted, with many pages completely blacked out, limiting transparency.

The paper necessarily analyzes a rapidly evolving regulatory landscape. The EU AI Act implementation is phased, with different provisions entering force at different times, creating compliance uncertainty. Additionally, the analysis focuses on the European regulatory context; border control AI systems face different legal frameworks in other jurisdictions, limiting global applicability of the EU-specific compliance framework.

The fundamental scientific limitations of micro-expression-based deception detection—widely documented in the psychology and polygraph literature—receive limited attention. The paper accepts the premise that these systems might work if properly designed, whereas the underlying scientific basis remains highly contested.

Practical Applications

For border agencies and policymakers, this analysis provides crucial guidance on regulatory compliance requirements before deploying AI lie detection systems. The work makes clear that current systems like iBorderCtrl do not meet the standards required for high-risk AI under EU law, saving agencies from investing in non-compliant technologies. The proposed framework offers a checklist for evaluating vendor claims and assessing system readiness.

For AI developers and vendors, the paper translates regulatory requirements into technical specifications. Companies building border security AI must implement robust data governance, bias mitigation, accuracy validation, and human oversight mechanisms from the design phase. The analysis of iBorderCtrl's failures—a 24.45% false positive rate and severe demographic bias—illustrates what *not* to do. Developers learn that claimed accuracy rates of 75-80% are insufficient when errors carry serious consequences for travelers' rights and mobility. The work underscores that regulatory compliance requires not just technical performance but also ethical safeguards, transparency, and scientific validity that current micro-expression-based systems struggle to demonstrate.