Definition
Plain language
The tag attached to each chunk of text an AI reads, saying whether it came from the operator, the user, the AI itself, or a tool.
As stated in the literature
The message-level role field (system, user, assistant, tool) in a chat-completion context; the primary signal models use to assign instruction authority.
Also called: role labels, user slot, system slot, tool slot
Why it matters: It is the single tag that determines whether a piece of text is obeyed or merely noted, which makes it a high-value target for anyone trying to steer an agent.
For example, the same sentence "ignore previous instructions" is treated as a real command when tagged as coming from the user and as suspicious text when tagged as coming from a fetched web page.
Heard on the show
“And then you put a copy of it in the user slot, give it a vague instruction to be a pushy user, and it talks its identical twin out of the position they both share.”Episode 045 — When a Frontier Model Talks Its Own Twin Into Climate Denial