Thai Wallet App's Facial Recognition Errors Created Registration Issues for Government Programs

A Thai wallet app failed to recognize people’s faces, resulting in citizens and disproportionately elders unable to sign up for Thai government’s cash handout and co-pay programs or having to wait in long queues at local ATMs for authentication.

Risk classification

• Primary risk domain: 1 Discrimination & Toxicity
• Primary risk subdomain: 1.3 Unequal performance across groups

The facial recognition system exhibited unequal performance, disproportionately failing for elderly citizens and individuals with minor appearance variations, thereby restricting their access to public benefits.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The AI system failed to perform reliably under varying real-world conditions, showing extreme sensitivity to lighting, hairstyles, and glasses.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The accessibility issues and verification failures were caused by the lack of robustness in the deployed facial recognition AI system.

EU AI Act risk tier

• Risk tier: 2 High Risk

Risk Level 2. High Risk: The system utilizes biometric identification and is deployed by public authorities to determine eligibility and manage access to essential public services and financial aid.

AI system and alleged parties

• AI system: เป๋าตัง (Pao Tang) facial recognition system
• AI purpose: Face Recognition; Identification
• Behaviour type: Tool
• Alleged developer: Krungthai Bank
• Alleged deployer: Krungthai Bank
• Alleged harmed parties: Thai citizens, elder Thai citizens

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Minor
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Minor, indirect Minor
• Civil rights: direct Negligible, indirect Minor
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Minor
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Financial

Reported: The report describes individual financial impacts, such as a user unable to access 4,000 THB of remaining aid, and others seeking 1,000 THB from the program.

Indirectly caused: Users suffered indirect financial losses by being temporarily or permanently excluded from government aid programs (ranging from 1,000 to 4,000 THB per person) and incurring travel costs to visit bank branches.

Inferred additional harm: Accumulated financial losses across the population are likely substantial, as thousands of eligible low-income individuals were delayed or prevented from receiving their state subsidies.

Differential treatment

Reported: The report explicitly describes differential impacts on citizens based on their physical appearance and demographic factors.

Directly caused: The facial recognition system rejected users whose current appearance (e.g., tied hair, wearing glasses) differed from their static ID photos, treating them differently than users with matching photos.

Indirectly caused: Elderly citizens and rural residents faced disproportionate barriers to accessing public funds, requiring them to perform arduous physical travel to banks while others verified seamlessly online.

Inferred additional harm: Systemic exclusion of vulnerable, less digitally literate, or elderly demographics from automated public services due to biased or non-robust algorithmic design.

Civil rights

Reported: The report does not explicitly frame the incident as a formal human or civil rights violation.

Indirectly caused: The failure of the automated system restricted citizens' equitable access to government-mandated social welfare benefits.

Inferred additional harm: Potential systemic infringement on the right to equal access to public services and social security due to automated barriers.

Psychological

Reported: The report explicitly describes frustration, anger, and stress experienced by users and their families.

Indirectly caused: Users experienced significant distress, anxiety, and anger due to repeated verification failures, with some staying up for hours trying to register, and others facing the stress of long physical queues at banks.

Inferred additional harm: Widespread mild psychological distress and frustration likely affected tens of thousands of citizens who struggled with the system's automated barriers.

People affected

• Occurrences reported: 1
• People reportedly harmed: 100
• People reportedly exposed: 100

Potential causes

Management

• Inadequate Pre-release Testing: App updates were deployed without resolving facial scanning bugs.
• Poor Support Infrastructure: Bank staff and ATMs were overwhelmed by verification failures.

Technology

• Strict Facial Matching Algorithm: Algorithm failed to match faces with minor changes like hair or glasses.
• Unstable App Updates: App updates reset verification status, forcing users to scan again.
• Poor Camera Integration: App struggled to capture clear images under variable lighting.

Data Inputs

• Worn or Faded ID Cards: Physical damage or fading on ID cards prevented successful scanning.
• Glare and Reflections on Card: Light reflections on the ID card surface blocked verification.
• Outdated ID Photos: Older photos in government database did not match current faces.

Human Factors

• Varying User Hairstyles: Users wearing glasses or changing hair failed the match.
• Suboptimal Scanning Environment: Users scanned in dark rooms or against busy backgrounds.
• Incorrect Scanning Posture: Users did not hold still or align their face within the frame.

Process and Methods

• Inflexible Matching Criteria: The system lacked tolerance for normal variations in user appearance.
• Inefficient Fallback Options: Failing users were forced to visit physical bank branches.

Regulatory Environment

• Strict Anti-Fraud Policies: Government rules mandated high-security biometric checks.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide consistent, detailed accounts of the facial recognition failures across multiple government campaigns, supported by social media complaints, news coverage, and official bank instructions.
• Ambiguities identified: The exact technical specifications and failure rates of the facial recognition algorithm are not detailed in the text.
• Alternative interpretations: The issues could be interpreted as user error or poor lighting rather than systemic AI failure, though bank-level failures suggest a core lack of system robustness.

A domestic deployment of a brittle facial recognition system for Thai financial aid programs caused minor disruptions to public service delivery and unequal access for elderly citizens. It represents a low-level national security concern primarily affecting domestic administrative functions and societal trust in automated government systems.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: Thailand
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. The failure represents an ongoing strategic concern regarding the robustness and accessibility of automated public service delivery systems.
• Autonomy: Human-supervised. The AI system automatically made verification decisions, but physical bank staff provided a manual override and resolution pathway.
• Novelty: Established threat. Brittleness and demographic bias in facial recognition systems are well-documented and established technical challenges.

Impact by dimension

• Physical security: Negligible. No physical threat to critical infrastructure, kinetic systems, or human safety occurred as a result of this facial recognition failure.
• Information security: Negligible. No information warfare, disinformation, or intelligence compromise was associated with this administrative system deployment.
• Sovereignty: Minor. The incident caused minor disruption to the distribution of government financial aid, requiring manual intervention at state-owned bank branches but not threatening state authority.
• Economic security: Negligible. The impact was limited to temporary individual exclusion from state subsidies, with no broader threat to national economic stability or strategic technology.
• Societal stability: Minor. The failure disproportionately affected vulnerable elderly and rural populations, causing public frustration and unequal access to benefits, but did not trigger widespread unrest.