Britain’s Financial Ombudsman Service (FOS) has found a growing number of AI hallucinations in consumer complaints to the body, leading it to call more people so they can discuss their case over the phone, the Financial Times has reported.

Chief ombudsman James Dipple-Johnstone told the paper that the most striking case of AI hallucination he had seen thus far had been when 37 out of 39 references made in a submission turned out to be incorrect.

Usage of AI in complaints can lead the substance of an individual’s submission to be lost in-between inaccurate, or hallucinated, AI material, Dipple-Johnstone said. He told the FT that some AI-written complaints included references that “weren’t relevant to the facts of this case, so you are having to go through that, but it generates extra effort for us in terms of having to work through those kinds of matters”.

The FOS is responding to this by calling more customers to discuss their complaints “to make sure that we get to the heart of their dispute,” Dipple-Johnstone added.

An FOS review of a small sample of cases from March found that up to a third of responses to its initial assessments were generated or heavily assisted by AI.

In a blog on its website, FOS chief operating officer Marc Harris noted that while AI could be “incredibly beneficial” to submissions when used well, excessive or uncritical use of the technology may lead to submissions that are excessively long, incoherent or simply incorrect.

This issue is not limited to individual’s submissions. In some cases, Harris said, claims management companies and legal professional representatives are using the technology to generate submissions on consumer’s behalf. In the most extreme examples, these have exceeded 200 pages and contained many errors or misunderstandings, he continued.

---