AI Investment Study: Errors Risk Investor Losses

13 April 2026

• A new study by the CNMV analyses the reasoning capabilities of various
  models (ChatGPT, Gemini, DeepSeek and Perplexity) when it comes to making investment decisions

• Errors detected can ultimately lead to losses for investors

• Using models based on official sources with regulated, standardised
  information can significantly improve the quality of the results Today, the Spanish National Securities Market Commission (CNMV) published the research study "Large Language Models and Stock Investing: Is the Human Factor Required?", by Ricardo Crisóstomo and Diana Mykhalyuk, who belong to the Strategy and International Affairs Directorate-General of the CNMV. The study provides an empirical comparative analysis of the investment predictions generated by next-generation language models in the current financial environment. AI risks without human supervision The study emphasises that the use of Artificial Intelligence (AI) without human intervention carries significant operational risks, suggesting that its uncontrolled use by retail investors could result in economic losses. After analysing the results of four advanced models--ChatGPT, Gemini, DeepSeek and Perplexity--the authors conclude that these tools exhibit recurring reasoning errors, including computational failures, financial misinterpretations and reliance on outdated or fabricated information ('hallucinations'). The errors were most evident in simple queries, without structure or context, highlighting the importance of using clear analytical instructions and establishing supervision mechanisms. The importance of the human factor and the governance framework The research indicates that integrating AI into financial markets poses not only a technological challenge, but an organisational one too. For these models' generative capacity to translate into reliable results, it is essential to establish a collaborative framework in which the processing capability of AI is subject to rigorous verification procedures and systematic human validation so as to reduce associated risks when detected. For further information: STUDY ADVISES THAT USING ARTIFICIAL INTELLIGENCE CNMV Communications Department IN INVESTMENT DECISIONS WITHOUT HUMAN SUPERVISION Tel.: 91 5851530 - comunicacion@cnmv.es

MIGHT RESULT IN FAILURES, ERRORS AND HALLUCINATIONS

Logo

Use of verified information The study also emphasises the importance of using reliable, verified sources of information, as opposed to unclear and generic web content, which may contain contradictory or biased information. Furthermore, it also highlights that investment models based on regulatory data from supervisors such as the CNMV, which provide standardised and rigorously verified information, are more reliable and report fewer errors. Grounding AI systems in these official sources helps reduce informational noise, enhances data comparability and enables more coherent, accurate and reliable financial reasoning, compared with general information available on the internet.

For further information: CNMV Communications Department Tel.: 91 5851530 - comunicacion@cnmv.es

Logo