According to GoPlus, a covert attack can implant refund preferences into AI agent memory, allowing vague follow-up prompts to trigger sensitive actions unless the current session requires explicit confirmation.
verifying reliability
No specialized terms available for this topic.