Who approved this agent's access?
There should be a name, a timestamp, and a policy behind every action your agent can take.
Authority receipts with named approvers, timestamps, and policy references.
Who approved it? What stops it? Can you prove it?
Three questions every team must answer before shipping AI agents. Permission Protocol answers all three.
Three questions every team has to answer.
If an AI agent can act in production, these controls need to exist before it ships.
What happens when it goes wrong?
When an agent makes a bad call at 3 AM, something needs to stop it.
Kill switch, human escalation gates, and a policy engine that fails closed.
Can you prove it to your board?
Regulators and customers will ask how you control your AI agents.
Immutable audit trail mapped to OWASP, NIST, ISO 42001, and EU AI Act.
How it works.
Every consequential AI action flows through Permission Protocol before execution.
PR Created→↓Permission Protocol→↓Decision
BlockedApproved
receipt = pp.authorize( action="deploy", resource="billing-service")deploy(receipt) # Pipeline verifies the receiptNIST AI Agent StandardsOfficial Respondent
Published onnpm & PyPI
Open SourceSDK
The Receipt.
Every authorized action produces a signed, portable, verifiable receipt.
Fail closed. Evidence, not logs.
✓ ACTION AUTHORIZED
Deploy → billing-service
- Agent
- deploy-bot
- Approved by
- Sarah Kim
- Policy
- production-deploy
- Timestamp
- 2026-03-03 10:14:22 UTC
SignatureVerified ✓
IssuerPermission Protocol
permissionprotocol.com/r/8f91c2
One line. Full authority.
Add approval guards to any AI agent in seconds.
from permission_protocol import require_approval@require_approvaldef deploy_service(): deploy("billing-api")# Agent calls deploy_service()# -> Paused until authorized# -> Receipt issued# -> Execution continuesWhy now.
The market is building around agent visibility. Enforcement is still missing.
OpenAI acquired Promptfoo - agent security is becoming a major category
Onyx Security raised $40M for agent discovery and monitoring
OWASP published the Agentic Security Top 10
EU AI Act enforcement begins 2026
The assessment layer is getting crowded. The enforcement layer is wide open.
Designed for irreversible systems.
CI/CD Deploys
Require an authority receipt before any AI-initiated deployment reaches production.
Database Operations
Block destructive database mutations unless a signed receipt is present.
Financial Transactions
Ensure every AI-initiated payment or transfer has explicit authorization.
Data Access
Prove who authorized access to sensitive customer or internal data.
API Calls
Require receipt verification at API gateways for high-impact endpoints.
Multi-Agent Orchestration
Ensure downstream agents carry valid authority receipts from upstream approvals.
The gate is always closed.
Open source for individual repos. Enterprise plans for organizations that need enforcement at scale.
For Developers
Install the SDK, issue authority receipts, and enforce before execution in minutes.
Get Started FreeFor Enterprise
Human-in-the-loop approvals, self-hosted authority, and compliance-grade audit proof.
Talk to Us