Skip to main content
    Governance & Risk

    Prompt injection

    What is prompt injection?

    Prompt injection is an attack in which malicious instructions hidden in content the model reads — a web page, an email, a document — cause it to ignore its original task or misuse its tools.

    The defence is architectural: treat all retrieved content as untrusted data, keep privileged actions behind explicit allowlists, and never let model output alone authorise a sensitive operation.

    Last reviewed by the gAIcko editorial team.