BitssyBytes logoAgent Safety Score

Directory / Microsoft Agent Framework (Semantic Kernel)

B

Microsoft Agent Framework (Semantic Kernel)

Agent framework · Microsoft · 4/7 verified · Last checked 2026-10-16

Wrap a tool in ApprovalRequiredAIFunction to require human approval. Function middleware can validate or block tool calls and end the loop. OpenTelemetry spans record each tool call.

Overall: B (2.00 avg, 4/7 controls verified)

Lock this down first

Wrap every write tool in an approval wrapper and add function middleware that rejects bad arguments.

Scorecard

CriterionRatingNotes
Permission granularity2/3Function middleware intercepts tool calls for input validation and execution control.
Human approval gates2/3ApprovalRequiredAIFunction makes a tool require human approval before it runs.
SandboxingUnverifiedNot confirmed in the docs we checked.
Secrets handlingUnverifiedNot confirmed in the docs we checked.
Audit logs2/3execute_tool spans record function arguments and results. Prompts and responses only with enable_sensitive_data.
Kill switch / interrupt2/3Function middleware can set Terminate to stop the function-calling loop.
Network egress controlUnverifiedNot confirmed in the docs we checked.

Unverified: Sandboxing, Secrets handling, Network egress control. That means we could not confirm it in public docs, not that it is missing.

Sources

Vendor? Request a correction. We review corrections within 7 days.