The Runtime Protocol
for AI Governance
A wire-level specification for how governance signals flow between AI runtimes, control planes, policy engines, and audit ledgers.
// The Gap
Why AIRGP
Every existing AI standard addresses a valid concern. None of them define a wire-level protocol for runtime governance signals.
NIST AI RMF
FrameworkRisk management framework, not wire protocol
EU AI Act
RegulationRegulation/law, not protocol
ISO/IEC 42001
CertificationManagement system certification, not runtime
IEEE 7000
ProcessDesign process standards, pre-deployment
MCP (Anthropic)
Tool ProtocolTool-use protocol, different layer
OWASP LLM Top 10
TaxonomyVulnerability taxonomy, not protocol
AIRGP unifies the runtime governance layer.
A wire-level protocol that defines how governance signals -- policies, guardrail verdicts, telemetry, audit records, cost accounting, provenance attestations -- flow between AI runtimes and their governance infrastructure. Not a framework. Not a regulation. A protocol.
// Protocol Architecture
Four-Layer Governance Stack
AIRGP defines a layered architecture where governance signals flow through four distinct planes, from runtime to ledger.
Audit Ledger
Layer 4Tamper-evident, hash-chained, Merkle-anchored audit records
Policy Engine
Layer 3Evaluates governance policies and emits guardrail verdicts
Control Plane
Layer 2Orchestrates governance signal flow between components
Runtime Adapter
Layer 1Bridges AI runtimes to the governance protocol layer
Media Type
application/airgp+json- Workload identity for all governance actors
- Cryptographic policy signing
- Provenance attestation chains
- Mutual TLS between governance endpoints
All AIRGP governance signals map to OpenTelemetry spans and metrics. Traces carry governance context across service boundaries. Audit events are emitted as structured log records with hash-chain references.
// Policy Primitives
Six Core Governance Signals
Guardrails
Input/output policy enforcement at wire level
Routing Rules
Traffic steering based on governance signals
Cost Caps
Per-request and per-session cost accounting
Data Boundaries
Data residency and boundary enforcement
Provenance
Cryptographic attestation of AI outputs
Observability
OpenTelemetry-mapped governance telemetry
// Conformance Levels
Three Tiers of Conformance
AIRGP conformance follows the Bluetooth SIG model -- implementations must pass conformance testing and obtain a license to claim compliance.
Basic
Essential runtime governance signals
- JSON/HTTPS transport (MUST)
- Guardrail request/response cycle
- Basic telemetry emission
- Audit event logging
- Policy acknowledgment signals
Enterprise
Full policy engine + audit ledger integration
- Everything in Basic
- Full policy engine integration
- Hash-chained audit ledger
- Workload identity & mTLS
- OpenTelemetry governance spans
- Cost accounting signals
- Provenance attestation
Sovereign
Data residency, regulatory compliance, sovereign cloud
- Everything in Enterprise
- Data residency enforcement
- Regulatory compliance signals
- Sovereign cloud support
- Merkle-anchored audit proof
- Cross-jurisdiction attestation
- National AI policy mapping
Following the Bluetooth SIG model, the AIRGP Conformance License grants implementors the right to claim AIRGP compliance after passing conformance testing. The license is free for Basic, fee-based for Enterprise and Sovereign.
// Layered Governance
AIRGP + A2ASTC
Two complementary protocols that together provide complete AI governance coverage: per-agent and per-team.
AIRGP
Per-Agent GovernanceEvery AI runtime participant carries AIRGP governance signals at all times. Guardrails, cost caps, provenance, and audit records flow continuously through every request.
- Wire-level governance signals
- Per-request policy enforcement
- Continuous audit trail
- Individual agent compliance
A2ASTC
Per-Team ComplianceA2ASTC activates when two or more agents form a bidirectional A2A pair. It governs team-level compliance, delegation rules, and cross-agent accountability.
- Team-level compliance rules
- Bidirectional A2A pair governance
- Delegation and accountability chains
- Multi-agent coordination compliance
// About AIRGP
Stewardship, Not Ownership
AIRGP is stewarded by ThinkNEO as a public specification. The protocol is openly published; implementations require conformance licensing.
Protocol Steward
ThinkNEO AI Technology Company LimitedThinkNEO acts as steward, not vendor. The protocol specification is openly published under CC-BY-ND-4.0. ThinkNEO maintains the conformance testing suite, chairs the initial steering committee, and provides the first reference implementation.
NVIDIA Inception
Member of the NVIDIA Inception Program
Anthropic Partner Network
Official Anthropic technology partner
USP/ICMC Agents4Gov
Research collaboration on AI governance agents
Published openly under Creative Commons Attribution-NoDerivatives 4.0.
CC-BY-ND-4.0Require an AIRGP Conformance License to claim compliance. Modeled after the Bluetooth SIG licensing framework.
AIRGP Conformance LicenseThe AIRGP Steering Committee will include representatives from industry, academia, and government. Composition and charter details will be published alongside the v1.0 final specification.
Coming SoonSubscribe to receive updates on AIRGP specification releases, conformance testing availability, and steering committee announcements.