Skip to content

Security & Privacy

Memory poisoning

Planting false or harmful information in an AI's saved memory to influence what it does later.

Example

An AI saves a misleading instruction as if it were a trusted preference.

Why people use it

It highlights attacks that plant misleading information for an AI to reuse later.

What you'll hear

“That saved preference didn't come from the user.”

What this means for you

Make saved memory inspectable and allow suspicious entries to be removed.

Can you control it?

No

No direct control. This describes a wider issue, concept or result rather than something you can simply switch on or off in a tool.

Common questions

Is memory poisoning limited to the current conversation?
No. Persistence can affect later tasks if the stored information is reused.
Can a harmless-looking note influence a later task?
Yes. Stored material may be treated as trusted background information when it is reused.
Is deleting the note always enough to undo the harm?
No. Actions already taken because of it may still need correction.

Related terms

Still have questions?

Up to 500 characters.

Ask LATHIC about AI. Relevant glossary entries may be included.

Your question, the glossary entries it matches, and a rotating pseudonymous identifier go to Microsoft Azure’s OpenAI service through Vercel AI Gateway to generate an answer. Zero retention and no training are required of the provider, and LATHIC does not save your question or answer. Privacy Notice