AI Recommendation Poisoning: When Optimization Becomes Manipulation

Microsoft Threat Intelligence Podcast episode summarizing research on "memory poisoning" (recommendation poisoning), a technique where websites embed hidden instructions in "Summarize with AI" links to manipulate AI assistants' memory and long-term recommendations; the episode explains differences from prompt injection, why businesses are adopting such tactics, potential long-term bias risks, and guidance for defenders and threat hunters on telemetry to monitor.