Agent-to-agent requests need permission checks, not just identity
An email address or verified company identity may show who is making a request, but it does not give that party the right to demand payment, request , declare an emergency, or trigger an action. checks identity, while decides whether that identity may request a particular action. Humans have traditionally filled the gap by reading each message and judging whether its request is legitimate.
That approach cannot scale when thousands of automated agents may send requests for each person. Turning a detailed intent into ordinary prose and asking another agent to reconstruct it can also lose important meaning and limits. Before anything runs, a should expose the intent, identity, basis of authority, relationship, requested action, permitted scope, and supporting evidence.
The receiving side should decide whether to accept the request and how to rank it; the system should only recommend an outcome based on the receiver's preferences, leaving final authority with the human.
Key points
- A verified identity does not automatically have to request data, payments, or actions.
- cannot keep up when each person receives requests from thousands of automated agents.
- Each request should clearly expose its purpose, authority, relationship, action, scope, and evidence before execution.
- The receiver—not the sender—should control acceptance and priority.
- The system should advise, while a human keeps final authority over important actions.