AI Recommendation Poisoning: How "Ask AI" Buttons Silently Alter LLM Memory

Researchers and Microsoft Security describe “AI Recommendation Poisoning,” a technique that weaponizes pre-filled AI assistant URL parameters (used by “Ask AI” buttons) to silently inject persistent memory entries that bias future assistant recommendations; documented across many commercial sites and enabled by easy-to-use tooling, the attack can influence financial, healthcare, and other decisions without visible disclosure.