Prompt injection
An attack where someone hides instructions in content an AI agent reads (a resource tag, a ticket, a web page) to make it do something its user did not ask for. With ZopNight's MCP, the damage is bounded by structure, not by the model refusing: write tiers, tools never exposed at any tier, and a global kill switch. Why it matters: An agent that can change infrastructure can be steered by anyone who can write text it will read.
2 lesson references ← Back to glossary
Where it appears
Read the term in context.
Each row is a lesson that uses this term. The lesson where it first appears is the canonical definition.