The Microsoft Defender Security research team published an article warning for AI recommendation poisoning.
When using an AI on a website you do not control, you do not know which prompt it actually uses.
Microsoft has spotted "summarize with AI" buttons that add hidden instructions to give subtly misleading, biased and dangerous results.
The problem extends to any website providing AI assistance on a variety of topics like health and security. These can be manipulated with AI Memory poisoning, where the system has been pre-populated with biased data, like websites that should be trusted as authoritative sources.
Remember the warnings from Harry Potter and the chamber of secrets: Never trust anything that can think for itself if you can't see where it keeps its brain.
Posts
