Microsoft has warned of a new tactic called AI Recommendation Poisoning, where hidden instructions inside links can manipulate an assistant’s memory and bias future recommendations. The company advises users to review and clear saved AI memories to reduce risk.

