What is it?
Memory poisoning (and prompt injection)
A security risk where malicious or misleading instructions get written into an agent's memory or context, and the agent later treats them as legitimate. "Indirect prompt injection" is when the harmful instruction is hidden in content the agent reads (a web page, a document) rather than typed by the user. Because memory persists, a single poisoned entry can corrupt the agent's behavior for a long time.
From the AI agent glossary