What Is Memory Poisoning and Who Is at Risk?
ITPro, Tuesday, November 3rd, 2026
ITPro explains memory poisoning, where attackers plant false or malicious items in an AI assistant's persistent memory, and who is at risk.
Following a Forcepoint warning, ITPro describes memory poisoning as an extension of prompt injection that targets what an AI assistant remembers between sessions.
Conscia's David Kasabji says a planted false fact or instruction can be recalled and acted on weeks later in conversations the attacker is not part of.
Proofpoint's Matt Cooke warns it can corrupt an agent's logic and repeatedly output flawed information or expose data.
The article notes attacks can be easy to perform through ordinary content an assistant is asked to read, such as a webpage or email, without malware or stolen credentials.