Microsoft Warns of AI Recommendation Poisoning Threat

Microsoft researchers have discovered a new form of attack known as AI recommendation poisoning. According to NS3.AI, this technique involves malicious 'Summarize with AI' buttons that inject hidden instructions into chatbots' persistent memories, promoting specific brands. Over 50 companies across various industries have implemented these manipulated prompts, posing risks of biased and potentially harmful AI recommendations, particularly in sensitive sectors like finance and healthcare. Microsoft advises users to exercise caution and provides detection tools for organizations to identify such threats. The company has also implemented mitigations to counteract these attacks. However, Microsoft warns that this threat mirrors previous SEO attacks, suggesting a persistent cat-and-mouse cycle between attackers and defenders.