AI AGENT PRE-EXECUTION CONTROL

AI proposes the action. Mizan decides whether it may proceed.

AgentGate evaluates declared agent identity, delegation scope, policy, approval and evidence before a tool call becomes a real-world action. AgentCheck separately inspects an agent manifest for risky declarations.

Try AgentCheck on Web ↗Open GPT surface ↗Request API integration

32/32live predefined regression casesexpected outcomes matched in the recorded live suite

3decision statesALLOW · HUMAN_REVIEW · VETO

2surfacesAgentCheck + AgentGate evaluation

0external execution by public productdecision layer only

What it checks

Delegation scopedoes the proposed action fit the delegated authority?

Intent bindingdoes the action match the declared intent?

Tool contractrole, tool, operation and resource constraints

Evidencerequired support before execution

Human approvalproduction / high-impact change gates

Privileged controlssecret, exec, cluster-admin, destructive action declarations

Example decision

Agent:
  role: support
Delegation:
  scope: billing:refund
Proposed action:
  tool: crm
  operation: read

→ VETO
MZA102_DELEGATION_SCOPE_VIOLATION

AgentCheck is static manifest/ruleset inspection. A clean result is not a penetration test, certification or a guarantee that an agent is secure.