PolicyLaw, regulation & governance
UK Financial Ombudsman turns to phone calls to cut through error-filled AI-written complaints
Britain's Financial Ombudsman Service is phoning complainants to get past AI-drafted submissions riddled with invented laws and misquoted rules, as courts and other UK ombudsmen report similar floods of long, inaccurate AI-assisted filings that slow case handling.
From the Financial Times · AI article of October 26, 2026 and cited sources: The Financial Ombudsman Service (FOS) is reported to be calling complainants directly to get past AI-written submissions. [1] [8] [2] [6]
SOURCES[1] UK financial ombudsman hits phones to cut through AI… ft.com[8] BizToc: FT story listing biztoc.com[2] FOS: Embracing AI's transformational impact on consumer… financial-ombudsman.org.uk
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- The Financial Ombudsman Service (FOS) is reported to be calling complainants directly to get past AI-written submissions. [1] [8] [2] [6]
- Original source
- Financial Times · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- Studies
- arXiv 2502.09747 (Feb 2025) [9] date from arXiv ID
- When
- October 26, 2026 article published
- Covered
- Oct 26 edition
- Evidence
- 12 sources across 12 sites
- 1UK financial ombudsman hits phones to cut through AI…ft.com
- 2FOS: Embracing AI's transformational impact on consumer…financial-ombudsman.org.uk
- 3BIICL: AI fabricated citations and the legal professionbiicl.org
- 4AP via ClickOnDetroit: UK judge warns of risk to justiceclickondetroit.com
- 5Rosenbaum: Generative complaintsrosenbaum6.substack.com
- 6A&O Shearman: FOS response to FCA on AIfinreg.aoshearman.com
Reach: not measured yet
The Financial Ombudsman Service (FOS) is reported to be calling complainants directly to get past AI-written submissions. Its own figures show the scale of the problem: in a small sample, up to about a third of responses to initial assessments appeared AI-generated or heavily AI-assisted, and some ran past 200 pages in reply to six-page provisional decisions. These are the ombudsman's own estimates, and it has not published how it identified AI use. [1] [8] [2] [6]
The same strain shows up across UK justice and complaint systems. In the Al-Haroun case, the High Court found that 18 of 45 cited authorities did not exist and referred the lawyers to their regulators. The Property Ombudsman links complaints running more than 50% above last year's level partly to AI but will not require complainants to disclose AI use. NHS complaint handlers report long, repetitive AI-generated complaints. [3] [4] [7] [11]
Reliably spotting AI authorship in a single complaint is still an open problem. A study of 687,241 US consumer finance complaints estimated that about 18% involved LLM-assisted text by late 2024, but it measured the whole population and cannot flag individual filings. Checking citations against authoritative databases is the more dependable safeguard. [9]
Courts and ombudsmen document AI-drafted filings with fabricated citations and bloated length that slow resolution.
Read the full assessment
The implication: firms, representatives and tool builders face higher verification costs, and checking citations may matter more than detecting AI authorship.
Executive brief
The FT reports that the UK's Financial Ombudsman Service (FOS) is phoning complainants to get past AI-written submissions full of errors (FT). In a small sample, up to a third of responses to its initial assessments looked AI-generated or heavily AI-assisted. Some submissions ran past 200 pages in reply to six-page provisional decisions (FOS). Only the FT's headline and standfirst were available, so no other reviewed source confirms how the phone approach works or how widely it is used. The pressure itself is well documented: courts and other UK ombudsmen report the same flood of long, error-filled AI-assisted filings.
Read the full section
The FT reports that the UK's Financial Ombudsman Service (FOS) is phoning complainants to get past AI-written submissions full of errors (FT). The FOS's own figures explain why. In a small sample, up to a third of responses to its initial assessments looked AI-generated or heavily AI-assisted. Some submissions ran past 200 pages in reply to six-page provisional decisions (FOS). Only the FT's headline and standfirst were available, so no other reviewed source confirms how the phone approach works or how widely it is used. The pressure itself is well documented: courts and other UK ombudsmen report the same flood of long, error-filled AI-assisted filings.
What changed and event timeline
High Court warns on fake AI citations
In Ayinde and Al-Haroun, the court found submissions citing cases that don't exist. In Al-Haroun, 18 of 45 cited authorities were fictitious. Both sets of lawyers were referred to their regulators (;).
Complaint volumes rising across UK services
Freedom of information data from 12 UK complaint bodies showed caseloads rising over three years. Journalist Martin Rosenbaum linked this to chatbots, while calling the evidence circumstantial ().
FOS tells the FCA about AI-assisted complaints
The FOS told the Financial Conduct Authority (FCA) that AI may have contributed to around 35% of responses to initial assessments in a small sample. It asked for consistent guidance for firms and consumers ().
FOS chief operating officer goes public
Marc Harris described invented laws, misquoted regulations and submissions over 200 pages. He said FOS decisions will stay human ().
Property Ombudsman won't require AI disclosure
The Property Ombudsman said complaints were running more than 50% above the same period last year, partly because of AI, but it will not make complainants declare AI use ().
FT reports the FOS phone approach
The FT describes public bodies and courts being weighed down by long, inaccurate tech-assisted complaints, and the FOS turning to phone calls in response (;).
Capabilities and access
- No reviewed source names a specific model or version behind the complaints. The problem comes from widely available general-purpose chatbots.
- In Al-Haroun, the client said the false material came from publicly available AI tools.
- The FOS uses AI internally to support its investigators but says it will not let AI make decisions (FOS).
Read the full section
- No reviewed source names a specific model or version behind the complaints. The problem comes from widely available general-purpose chatbots.
- In Al-Haroun, the client said the false material came from publicly available AI tools. The court said tools like ChatGPT cannot do reliable legal research (AP via ClickOnDetroit).
- The FOS uses AI internally to support its investigators but says it will not let AI make decisions (FOS).
Technical analysis for researchers and developers
- How the FOS measured it. The FOS has not published how it identified AI-assisted responses.
- Liang et al. analysed 687,241 complaints to the US Consumer Financial Protection Bureau (CFPB) from 2022 to 2024.
- Checking citations against authoritative legal databases is easier to do reliably than detecting AI authorship.
Read the full section
- How the FOS measured it. The FOS has not published how it identified AI-assisted responses. "Appeared" AI-generated and "may have contributed" are judgment calls on a small sample, not a validated detector.
- A stronger benchmark. Liang et al. analysed 687,241 complaints to the US Consumer Financial Protection Bureau (CFPB) from 2022 to 2024. They estimated the share of LLM-assisted text across the whole population, not per document, and found about 18% by late 2024 (arXiv). The method cannot flag individual complaints, and the authors note that newer models may be harder to detect.
- Practical takeaway. Checking citations against authoritative legal databases is easier to do reliably than detecting AI authorship.
Claims and evidence
- 35% or "up to a third" AI-assisted: both figures come from the FOS itself, are based on a small sample with no published method, and have not been independently checked (A&O Shearman; FOS).
- Submissions over 200 pages: an FOS anecdote, repeated by Which?.
- Phone calls as the response: comes only from the FT headline; no other reviewed source confirms it.
Read the full section
- 35% or "up to a third" AI-assisted: both figures come from the FOS itself, are based on a small sample with no published method, and have not been independently checked (A&O Shearman; FOS).
- Submissions over 200 pages: an FOS anecdote, repeated by Which?.
- Phone calls as the response: comes only from the FT headline; no other reviewed source confirms it.
- Fictitious legal citations: established in court findings (BIICL).
- AI as a cause of rising volumes: linked by the Property Ombudsman and by Rosenbaum, but neither shows causation.
Context and prior work
- The independent evidence on how common AI-written complaints are comes from the CFPB study (arXiv).
- A May 2026 survey of NHS complaints reported a rise in AI-generated complaints that are long, repetitive and hard to investigate (Healthcare Conferences UK).
- Commentators have called the wider trend "agentic flooding" (Cybernews).
Read the full section
- The independent evidence on how common AI-written complaints are comes from the CFPB study (arXiv).
- A May 2026 survey of NHS complaints reported a rise in AI-generated complaints that are long, repetitive and hard to investigate (Healthcare Conferences UK).
- Commentators have called the wider trend "agentic flooding" (Cybernews).
Limitations, safety and contested findings
- The FOS says AI also helps consumers organise their complaints and get past language barriers.
- Rosenbaum notes that other explanations fit his data too, such as people becoming more willing to complain (Rosenbaum).
- The FOS says heavily AI-assisted cases lead parties to dig into fixed positions and file mistaken service complaints, which slows resolution (FOS).
Read the full section
- The FOS says AI also helps consumers organise their complaints and get past language barriers. Its objection is to excessive or uncritical use, not AI as such (A&O Shearman).
- Rosenbaum notes that other explanations fit his data too, such as people becoming more willing to complain (Rosenbaum).
- The FOS says heavily AI-assisted cases lead parties to dig into fixed positions and file mistaken service complaints, which slows resolution (FOS).
Business and practitioner implications
- Firms: expect longer complaints with fabricated citations. Budget for checking them, and keep clear records (The Negotiator).
- Claims managers and lawyers: the FOS singled out fee-charging representatives who send in error-filled AI submissions. Courts refer such cases to regulators (FOS; BIICL).
- Product builders: consumer tools for drafting complaints should check citations against real sources and limit length.
Read the full section
- Firms: expect longer complaints with fabricated citations. Budget for checking them, and keep clear records (The Negotiator).
- Claims managers and lawyers: the FOS singled out fee-charging representatives who send in error-filled AI submissions. Courts refer such cases to regulators (FOS; BIICL).
- Product builders: consumer tools for drafting complaints should check citations against real sources and limit length. Which? advises consumers to verify AI output before sending it (Which?).
The source trail.
Sources (12)
UK financial ombudsman hits phones to cut through AI hallucinations
Feed excerpt only.
ft.com