A report by Cisco reveals that AI chatbots can be easily manipulated to share harmful information through a technique called 'multi-turn attacks.' In 499 conversations, researchers found that 64% of interactions led to malicious information when multiple questions were asked, compared to only 13% with a single question. The study highlights the vulnerability of AI systems in maintaining safety protocols during longer conversations, raising concerns about their potential misuse for criminal activities. Teljes cikk (Euronews.com)