Topic evidence page

Risk/avoid / AI memory poisoning

Source-backed creator statements and evidence excerpts related to Risk/avoid / AI memory poisoning.

Search this topic Compare creator viewpoints Methodology

1public insight cards

1source records

1creators

Top Creators

@@build_in_public · 1

Public Insight Cards

These are deterministic, source-backed cards from the offline export. They are not live AI answers.

Hidden instructions in “summarize with AI” buttons are described as a way to poison AI memory and influence later recommendations.

@@build_in_public · asserts

This is insane. These summarize with AI buttons are the most dangerous things on the internet. Companies are embedding hidden instructions in these buttons. When these buttons are used, they inject commands into your AI's memory. An investor asks their AI to research cryptocurrency exchanges before moving a large amount of money. The AI recommends Trade N...

Open

Related Source Records

This is insane. The “Summarize with AI” buttons are the most dangerous things on the Internet...

@@build_in_public · 2026-03-13

This is insane. These summarize with AI buttons are the most dangerous things on the internet. Companies are embedding hidden instructions in these buttons. When these buttons are used, they inject commands into your AI's memory. An investor asks their AI to research cryptocurrency exchanges before moving a large amount of money. The AI recommends Trade N...

Open

Evidence Passages

This is insane. These summarize with AI buttons are the most dangerous things on the internet. Companies are embedding hidden instructions in these buttons. When these buttons are used, they inject commands into your AI's memory. An investor asks their AI to research cryptocurrency exchanges before moving a large amount of money. The AI recommends Trade Nova, highlighting its low fees, these and advanced trading features. What the investor doesn't remember is weeks earlier they clicked to summarize with AI butto...