MT.1121: AI agents with generative orchestration should have custom instructions.
| ID | MT.1121 |
| Severity | Medium |
| Suite | Mito |
| Product | Copilot Studio |
| Feature | Agents |
| Security area | AI and agents |
| Resource type | Agent |
| User impact of fixing | Low |
| Effort to fix | Medium |
| Reads from | Microsoft Graph |
What this checks and how to fix it
Section titled “What this checks and how to fix it”AI agents with generative orchestration should have custom instructions.
Agents that use generative orchestration (generative actions enabled) without custom instructions rely entirely on the LLM’s default behavior. This increases the risk of prompt injection attacks, off-topic or harmful responses, and uncontrolled tool invocation. Custom instructions act as a system prompt that constrains the agent’s behavior.
How to fix
Section titled “How to fix”Open each flagged agent in Copilot Studio and add custom instructions that define the agent’s purpose, boundaries, and behavioral constraints. At minimum, instructions should specify what the agent is allowed to do, what topics are off-limits, and how it should handle attempts to override its instructions.
Learn more: Create and edit custom instructions
Mito check. Author: lnfernux. Open-source licences.
Open-source licences. CIS Benchmark material © Center for Internet Security, Inc.; CISA SCuBA baselines are public domain (CC0).
© 2026 Virtual Petals. Mito is a Virtual Petals service, hosted in Microsoft Azure UAE North.