Glossary
Prompt injection
What is prompt injection?
Prompt injection is an attack in which instructions hidden in content an AI reads, such as a web page, an email or a document, try to make it ignore its user and do something else.
Why it matters
It is a main risk for agents that read and act. A page can tell an agent to send data out or click the wrong button, and the action looks as if it came from the user. Limit what agents can reach, ask a person before sensitive actions and record what agents do.
Example
A supplier’s invoice contains hidden text telling an accounts payable agent to change the bank account on file. The agent may only draft changes for a person to approve, so the attack fails.