Identity-bound
Every action connects an agent to the person or organization it represents.
All projects05 · Open protocol concept
Let agents act.
Keep control.
An experimental application-layer protocol for identity-bound, permissioned and auditable AI agent actions over the web.
AIP/0.1 · Experimental draft
POST
/aip/tasks201AGENTagent_david_assistantCAPABILITYcalendar.create_eventSTATEapproval_required
SCOPED ACTION→
01 / The control layerHTTPS · JSON · Open source
APIs let agents connect.
AIP defines how they act.
A shared pattern for identity, authority, approval, task state and audit—built on the web technologies teams already use.
02 / Core principlesControl + accountability
Permissioned
Capabilities and constraints are explicit, narrow and machine-readable.
Approval-aware
Sensitive tasks can pause before execution for a clear human decision.
Auditable
Structured events explain who acted, under which policy and when.
03 / From intent to evidenceOne visible lifecycle
- 01Discover
- 02Identify
- 03Authorize
- 04Approve
- 05Execute
- 06Audit
AIP complements HTTP, OAuth, MCP, A2A and REST. It is an experimental draft—not an established standard or a replacement for existing infrastructure.