What is prompt injection?
Prompt injection is an attempt to trick an AI agent with instructions hidden in a message or document, such as telling it to ignore its rules or reveal private data.
Any agent that reads messages or documents from outside the business can be sent instructions disguised as content. The risk is the agent treating that content as orders.
The defense is layered: keep outside content separate from the agent's own instructions, limit what the agent can reach, and make sure there are no secrets in its context to reveal.
Ready to put your first agent into production?
Bring us one process. We’ll map the skills, rules and integrations, and help you choose cloud, on-premises or a managed service.