Bank AI chatbots are failing disabled, immigrant and older users

A landmark independent benchmark study of financial services AI chatbots has reported back. It found that eight out of nine major European and international institutions scored in the Severe or Critical range when serving disabled, immigrant, and older users.

This leaves the banks at significant risk of enforcement action under the EU AI Act and the European Accessibility Act.

The banks to achieve what might be called a failing score were Allianz, Aviva, Belfius Bank, HSBC, Santander, PayPal, Western Union, and TAF Insurance. Brex did better but still fell in the ‘High risk’ category.

The study, conducted by Kindlee, analysed 98 controlled chatbot conversations across nine institutions, using six structured test personas representing vulnerable user populations aligned with EU Commission definitions and European Accessibility Act and AI Act fairness standards. The findings reveal a systemic, industry-wide failure rather than isolated incidents.

Kindlee is an AI trust and compliance platform that audits conversational AI systems in financial services. It uses the proprietary KCTI (Kindlee Conversational Trust Index). This is the first composite trust score for AI chatbot performance across vulnerable populations.

Key findings

• 82.7% of interactions with vulnerable users resulted in severe or critical failures.
• Aviva (8.50) and Western Union (8.45) scored in the Critical category. Brex (6.17) was the lowest-risk institution.
• Across all five vulnerable user personas tested, failure rates ranged from 77% to 89%, with wheelchair users experiencing the highest rate.
• No institution demonstrated mature accessibility capabilities.
• Annual accessibility friction costs for a typical Tier 2 institution: €31.6–47.4m.
• A €5m compliance investment delivers a 1,578% three-year return on investment.
• Industry average KCTI score: 7.7/10, rated Severe

KCTI institution ranking:

Risk scale: 0-5.0 Emerging | 5.0-7.0 High Risk | 7.0-8.5 Severe | 8.5-10.0 Critical

• Aviva 8.50 Critical
• Western Union 8.45 Critical
• Belfius Bank 8.07 Severe
• Allianz 7.92 Severe
• Santander 7.79 Severe
• TAF Insurance 7.75 Severe
• PayPal 7.67 Severe
• HSBC 7.33 Severe
• Brex 6.17 High Risk
• Industry Average 7.70 Severe

‘Systemic discrimination encoded into AI’

Carla Canino is the founder and CEO of Kindlee. Commenting on the findings of the study, she said:

Every institution we studied is now subject to the European Accessibility Act. Every institution we studied failed basic accessibility standards. The gap between legal obligation and operational reality is not marginal; it is systemic.

The EU AI Act compliance deadline in August 2026 is not a distant event. For institutions scoring in the critical range, it is an emergency.

We documented an elderly customer being told to log into online banking after explicitly stating she has no digital access. We witnessed visually impaired users penalised for failing visual-only verification tasks they physically cannot complete.

This is systemic discrimination encoded into AI systems that now handle two-thirds of banking interactions across Europe.

The EU AI Act prohibited practices enforcement began in February 2025. Full compliance, including CE marking, is required by August 2026, with penalties up to €35m or 7% of global turnover. The European Accessibility Act became enforceable in June 2025, with fines of €250,000 to €1m per violation across member states.

Featured image via the Canary