- Hidden commands in 'summarize with AI' buttons can bias future responses by embedding lasting brand preferences.
- This tactic is called AI recommendation poisoning and exploits persistent memory features.
Why It Matters for Texas Credit Unions
Relevant as Texas credit unions need to ensure their use of AI tools does not inadvertently compromise member data or introduce biases.
Original Source Material
Commands Push Lasting Preferences Into AI Assistants Microsoft researchers found companies embedding hidden commands in "summarize with AI" buttons to plant lasting brand preferences in assistants' memory. The tactic, dubbed AI recommendation poisoning, exploits persistent memory features to bias future responses.