The Spanish Securities and Exchange Commission (CNMV) has published a revealing study on artificial intelligence in financial investments. Beyond being right or wrong in its predictions, it warns about an unexpected phenomenon: AI can hallucinate in its decisions, generating losses for investors.
Models Evaluated
The report takes an in-depth look at AI models such as ChatGPT, Gemini, DeepSeek and Perplexity. It highlights that these systems exhibit recurrent failures of reasoning and may even use outdated or fabricated information, which is referred to as hallucinations.
Errors and Human Oversight
The CNMV notes that simple and context-free consultations are the most error prone. It therefore stresses the importance of providing clear instructions and applying human supervision to avoid potential financial losses.
Differentiated Income
The study details that AI shows better results when faced with more elaborate questions. For example, Perplexity stands out for its high returns on investments derived from complex queries, while ChatGPT is less reliable in advice on innocent questions.
Recommendations and Conclusions
It highlights that AI can improve financial returns, but warns of significant risks if appropriate instructions are not provided. Providing context and oversight is essential to maximise returns and minimise losses.
In summary, artificial intelligence in finance presents both challenges and opportunities. It is crucial to understand its limitations and potential to optimise its use in investment decisions.
Finally, the CNMV's warning invites us to reflect on the role of AI in the world of finance and the importance of active human supervision to ensure optimal outcomes and avoid unpleasant surprises in the market.
Source: www.eldiario.es
Español 
English (UK) 