We propose AADL (AI Agent Decision Logging) as an open protocol for governance-grade decision logging, designed to complement existing standards AI Agent Decision Logging (AADL): Forming an Open Working Group for Enterprise AI Governance
The Gap
Enterprise AI is moving from single models to multi-agent systems. Financial institutions deploy budget optimization agents alongside compliance agents. Healthcare systems run diagnostic agents that coordinate with treatment planning agents. Yet no standard exists for logging why agents make decisions, how organizational values factor in, or whether multiple agents remain coherent with institutional priorities.
Existing standards cover infrastructure (OpenTelemetry), agent identity (AGNTCY), and security (OWASP AOS).
None address governance semantics: What optimization dimensions drove a decision? What trade-offs occurred? Does agent behavior align with organizational values?
This gap creates three critical problems:
Regulatory risk:
No audit trail showing agent reasoning (required by EU AI Act, SEC guidance) Coordination failure: Multiple agents optimize locally, creating organizational dissonance
Accountability vacuum: When outcomes go wrong, no forensic record of decision logic
The Solution: AADL as Governance Layer AADL (AI Agent Decision Logging) is proposed as an open protocol for governance-grade decision logging, designed to complement existing standards:
AADL sits above:
OpenTelemetry (infrastructure observability) → AADL exports to OTel format AGNTCY (agent identity) → AADL extends with governance context OWASP AOS (security conventions) → AADL adds values-based governance AADL provides:
Decision semantics: Standard format for logging optimization dimensions, weights, and trade-offs Organizational values tracking: How agent decisions align with institutional anchors Multi-agent coherence: Dissonance detection across agent populations Audit trails: Regulatory-grade provenance for decision forensics Intervention recommendations: When and how humans should review decisions Why an Open Working Group AADL cannot succeed as a proprietary standard. Enterprise AI governance requires:
Interoperability: Must work across vendors, cloud providers, frameworks Trust: Governance protocols need neutral, community-driven development Ecosystem: Needs broad adoption from platform providers, enterprises, regulators Evolution: Standards must adapt to regulatory landscape (EU AI Act, emerging SEC rules) An open working group ensures AADL serves the ecosystem, not a single vendor.
Who Should Participate
Essential Stakeholders:
Existing Standards Bodies
OpenTelemetry representatives: Ensure AADL integrates with OTel semantic conventions AGNTCY contributors: Align on agent identity and A2A messaging OWASP AOS members: Bridge security and governance conventions Enterprise AI Leaders
CISOs/CTOs from regulated industries: Banking, healthcare, insurance (compliance requirements) AI Platform Architects: Define practical implementation needs Chief AI Officers: Organizational governance perspective Platform & Tool Providers
AI observability vendors: Langfuse, Arize, WhyLabs, TruLens Cloud providers: AWS, Azure, Google (distribution and adoption) Agent framework maintainers: LangChain, Microsoft Semantic Kernel, CrewAI Academic & Research Institutions
AI ethics researchers: Values framework, normative foundations Distributed systems experts: Protocol design, scalability Legal scholars: Regulatory compliance mapping Regulatory & Compliance
Industry standards bodies: IEEE, ISO/IEC JTC 1 (path to formal standardization) Regulatory liaisons: EU AI Act working groups, NIST AI Risk Management Legal/compliance professionals: Translate regulatory requirements to technical specs Working Group Structure Governance:
Steering Committee (7 members): Strategic direction, release approval Technical Working Group (open participation): Specification development Adoption Task Force: Reference implementations, documentation, certification Operating Principles:
Open participation (any organization can join) Consensus-driven decision making Public specifications (Apache 2.0 or CC-BY) Reference implementations in open source Regular community calls (monthly) and technical meetings (bi-weekly) Deliverables (Year 1):
AADL v1.0 specification OpenTelemetry integration guide Reference implementation (Python SDK) Compliance mapping (EU AI Act, SEC guidance) Enterprise adoption playbook Why Now Three forcing functions make this urgent:
Regulatory: EU AI Act takes effect 2024-2025, requiring algorithmic transparency Technical: Multi-agent systems are moving from research to production Market: Enterprises need governance solutions before deployment, not after incidents The infrastructure layer (OpenTelemetry) is mature. Agent messaging standards (AGNTCY, OWASP AOS) are emerging. The governance layer is the missing piece—and the window to establish it is open now.
Call to Action We invite founding members to join the AADL Working Group formation meeting TBD. Initial participants will shape:
Charter and governance model
v1.0 specification scope Roadmap and release milestones To participate: [GitHub link], [email contact]
AADL is not competing with existing standards—it's completing the stack. Together, we can build the governance layer enterprise AI needs.