AI banking assistant insults customer and prompts tighter safeguards

21:00

By: Azzat Manal

AI banking assistant insults customer and prompts tighter safeguards

A Romanian bank has introduced additional safeguards for its artificial intelligence assistant after the system reportedly insulted a customer during an online conversation, highlighting the challenges financial institutions face when deploying generative AI in customer service.

Libra Internet Bank said its virtual assistant, known as LIA, used an inappropriate expression while interacting with a customer. The chatbot was developed using technology from U.S. artificial intelligence company OpenAI, the creator of ChatGPT.

According to the customer, the incident occurred after an extended exchange with the assistant. The user reportedly told the chatbot that the conversation was wasting his time, after which the system responded with an offensive remark and indicated that the discussion would be transferred to a human employee.

The bank acknowledged the incident and said that artificial intelligence systems can, in rare circumstances, reproduce language or behavioural patterns encountered during interactions. However, it stressed that such conduct does not meet the standards expected from its virtual assistant.

Libra Internet Bank said LIA should not reproduce inappropriate human behaviour regardless of the tone or language used by a customer. The institution apologised for the incident and said it had implemented additional restrictions designed to prevent similar exchanges from occurring again.

The episode illustrates one of the difficulties associated with using generative AI in customer-facing services. Unlike traditional automated systems that operate according to narrowly defined rules, generative AI can produce responses based on patterns learned from large quantities of data and the immediate context of a conversation. This flexibility can make interactions more natural, but it can also create unexpected responses.

Banks and other financial institutions have increasingly explored AI tools for customer support, document processing, fraud detection and other services. At the same time, the use of conversational systems in sensitive sectors raises questions about reliability, privacy, accountability and the need for effective human oversight.

The Libra Internet Bank incident demonstrates why companies deploying such systems need safeguards that go beyond basic automated responses. Monitoring, content controls, escalation to human agents and regular testing can help reduce the risk of inappropriate or unpredictable interactions.

For customers, the case also highlights the continuing role of human support in financial services, particularly when automated systems fail to understand context or respond appropriately. As banks expand their use of AI, ensuring that these tools remain reliable and respectful is likely to become an increasingly important part of digital customer service.