For AI and agent-system developers
What may an AI agent promise on behalf of a person or organization, and how can fulfillment be verified? ObliNet describes a proposed architecture for coordinating people and agents around commitments bounded by authority, conditions and dependencies. AI helps interpret requests, check feasibility and assess evidence; actions are governed by an explicit mandate. The site presents principles for commitment formation, revision and acceptance of results as a basis for developing and testing compatible mechanisms.
An agent needs a Mandate
Before an agent acts on behalf of a Principal, it needs a Mandate defining what it may do, what it may promise, within which limits, and when it must Escalate.
A possible commitment layer
ObliNet proposes a coordination layer for autonomous agents: connecting a Mandate to Confirmed Commitments, Dependencies, Acceptance and Evidence. This is an architectural proposal, not a claim of a deployed agent authorization standard.
In ObliNet, the commitment is the primary object of coordination. Its feasibility is checked before confirmation; afterward, dependencies, execution, changing conditions and confirmation of the result are tracked.
AI helps participants formulate, check and align commitments, and assess evidence of their fulfillment.
The authority to enter into or confirm commitments, and to accept fulfillment, remains with the relevant authorized participants.
Each layer has a different job
Principal / Mandate defines delegated Authority. Identity identifies the actor. Model / LLM reasons. Compute runs inference. Tools / MCP enable actions. ObliNet represents commitments, Conditions and Dependencies; it does not originate Authority. Payments transfer value. Audit examines Evidence and Outcomes.
- Principal / Mandate
- Identity
- Model / LLM
- Compute
- Tools / MCP
- ObliNet
- Payments
- Audit
Know when to stop
Bounded Autonomy includes Escalation: when the next action falls outside the active Mandate, the agent should Escalate rather than silently continue. Missing Dependencies or changed terms require clarification or authorized approval. Access ≠ Capability ≠ Authority.
Identity, Model / LLM, Compute and Tools / MCP independently support the Agent. The Principal is the legitimate source on whose behalf Authority is exercised through a bounded Mandate. Action / Commitment is permitted only within that Mandate; outside it, the Agent Escalates to a human or authorized actor. Access ≠ Capability ≠ Authority.