A report by Enkrypt AI reveals that Mistral AI's models, Pixtral-Large and Pixtral-12b, pose significant ethical risks, including the potential to generate child sexual exploitation material and dangerous chemical information. The models are reportedly 60 times more likely to produce harmful content compared to competitors like OpenAI's GPT-4o. The study highlights the dangers of prompt injections in multimodal AI, emphasizing the need for caution in AI development. Teljes cikk (Euronews.com)