Skip to content
Submit incident
AII-1000-69103
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.Supporting sources are pending verification.

Spain's market regulator warns AI systems make investment errors without human oversight

Spain's financial regulator CNMV published a study showing that AI systems like ChatGPT, Gemini, DeepSeek and Perplexity make errors and hallucinate when making investment recommendations, potentially causing investor losses without human oversight. The systems confused companies, provided false data, and mixed information from different periods, though performance improved with structured prompts and standardized data sources.

13 April 2026 路 Comisi贸n Nacional del Mercado de Valores 路 Reliability and hallucination
13 April 2026
AII-1000-69103
Quick facts
Category
Reliability and hallucination
Country
Spain
Date
13 April 2026
Drafted by
Raidu Incident Detection

What happened

The Spanish financial regulator CNMV published a study analyzing how ChatGPT, Gemini, DeepSeek and Perplexity make investment decisions about publicly traded companies. The study, conducted by researchers Ricardo Cris贸stomo and Diana Mykhalyuk, found that these AI systems exhibit errors and hallucinations that without human oversight can cause investors to lose money.

When analyzing stocks in the IBEX index, the AI systems confused companies within the same group, provided non-existent stock quotation data, mixed earnings figures from different years, and incorrectly categorized some securities. The researchers noted that AI performance improved significantly when given structured prompts and standardized data sources rather than unfiltered web information.

Reported impact

The CNMV concluded that errors detected in AI investment recommendations can ultimately lead to losses for investors. The regulator stressed the necessity of expert human supervision to mitigate these risks.

Index analysis

The researchers found that large language models can potentially achieve market-beating performance, but this conclusion is qualified by significant caveats related to how the AI is prompted and what data sources it uses. Performance varied substantially depending on whether users interacted with the AI in a naive or structured manner.

Classification

Category
Reliability and hallucination
Industry
Financial and insurance services
Country
Spain
Harm status
Potential harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Economic or property
Harmed
Consumers
Failure modes
Agent misaction, Hallucination, Misidentification, Unsafe recommendation
AI systems
ChatGPT, DeepSeek, Gemini, Perplexity

Sources and evidence

Sources not yet published

This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.

How sources are verifiedSuggest a source for this incident