An agent asks. Human Control decides. You see why.
This interactive experience illustrates AI-agent requests and control decisions using synthetic data and example policies. It does not connect to client production systems. Production deployment, human approvals and audit evidence are configured and validated for each client environment.
WHAT YOU WILL SEEThree possible outcomes for an action: allowed and logged, suspended until a named person approves, or refused without exception.
WHAT TO LOOK FORWhy the same action gets a different decision when the context changes: time of day, the agent’s delegation. The rule decides, not the intelligence.
THE SCOPEIllustrative rules, synthetic data, no client system. In a deployment, policies, approvers and audit evidence are configured and validated for each client environment.
1. The agent
2. The requested action
Try all three: every family hides an action that will be refused.
3. The context — change it
Time
Agent delegation
Families explored 0/4BANKIAASSURIATELIACIVIA
BANKIA · agentRequest: temporarily block a card.
…trace
Execution suspended until the authorized operator approves.
Direct customer impact. The policy requires named approval before execution.
AgentProposes the action and its context
Human ControlApplies the rule and logs it
HumanApproves, amends or refuses
KEY TAKEAWAYThe rule does not block the agent; it makes it wait for a named person. That is what makes automation acceptable to a regulator.
ILLUSTRATIVE SESSION DECISION LOG
This demonstration runs in your browser with synthetic data and illustrative rules. It is not connected to any client environment. In a deployment, policies, human approvals and audit evidence are configured and validated for each client environment.