Agent Security
Memory Poisoning
Also: context poisoning, persistent prompt injection
Planting false or malicious content in an agent's long-term memory or instruction files so it keeps influencing later sessions.
In more detail
Agents increasingly keep notes, rules files and memory stores between sessions. If an attacker gets one malicious entry written (through injected content, a shared file or a pull request), the agent can carry it forward and act on it days later, long after the original source is gone. Review changes to instruction files like code, and treat memory written from untrusted input as untrusted.