AI agents with generative orchestration should have custom instructions
Why This Matters
AI agents with generative orchestration can interpret user requests broadly, potentially deviating from your organization's intended scope or security boundaries. Without custom instructions, these agents might expose sensitive information, perform unauthorized actions, or fail to reject prompt injection attempts. Ensuring clear guidance is in place prevents misuse and maintains control over automated workflows.
What Aether365 Checks
Aether365 verifies whether each Copilot Studio agent using generative orchestration has custom instructions configured. This check appears in your Aether365 dashboard under the microsoft-365 service checks as rule AE.1121.