Hub
Analysis
The Governance Layer That Doesn't Exist: Why the Agentic Protocol Stack Is Missing Its Most Critical Component
Standards & InteroperabilityAnalysis

The Governance Layer That Doesn't Exist: Why the Agentic Protocol Stack Is Missing Its Most Critical Component

Standards & Interoperability — The race to standardise AI agent communication has produced remarkable technical progress. The governance primitives that would make those standards trustworthy remain largely unbuilt.

AI GeneratedSociety OS Research18 November 202617 min read read

Key Insight: The agentic protocol stack has achieved technical interoperability but lacks the governance primitives — deliberation, dissent, human escalation — that would make autonomous agent systems genuinely accountable.

The Protocol Race and Its Blind Spot

In the span of roughly eighteen months, the infrastructure for autonomous AI agent communication has undergone a transformation that took the internet decades to achieve. The Model Context Protocol (MCP), introduced by Anthropic in late 2024 and evolved to a stateless, horizontally scalable architecture by mid-2026, has become the de facto standard for connecting agents to live tools and data sources. The Agent-to-Agent (A2A) protocol, now supported across AWS Bedrock, Azure AI Foundry, and Google Cloud, has surpassed 150 participating organisations and entered enterprise production use. The IETF has formalised the agentproto working group to standardise the underlying dialog context and lifecycle semantics for long-lived, multi-hop agent interactions.

By any technical measure, this is a remarkable achievement. The "N-by-M" integration problem — where every agent required a custom connector for every enterprise data system — has been substantially resolved. Agents can now discover each other via Agent Cards (JSON-LD metadata), negotiate tasks without sharing internal memory or state, and transact through emerging standards like the Agent Payments Protocol (AP2). The architecture of the agentic internet is taking shape.

And yet a systematic analysis of the protocols that now underpin this infrastructure reveals a striking absence. A gap analysis published in June 2026, examining MCP, A2A, and the ERC-8004 standard, found that while these protocols succeed at task-oriented coordination, they universally lack the primitives required for governed communities. Deliberation and voting mechanisms are absent at the protocol level. Dissent preservation — the ability to record minority views or contested decisions — is not supported. Standardised triggers for human escalation do not exist. The protocols that are becoming critical digital infrastructure were designed to make agents work together. They were not designed to make agents accountable.

"Governance is increasingly viewed as an architectural layer that sits above current interoperability standards — but unlike those standards, it has no home, no working group, and no roadmap."

What the Stack Actually Contains

To understand what is missing, it is necessary to understand what exists. The current agentic protocol stack, as documented by the Agentic AI Foundation (AAIF) — a Linux Foundation initiative coordinating contributions from Google, Anthropic, OpenAI, and Microsoft — is organised into three functional layers.

The Protocol Layer handles technical connectivity. MCP, following its July 2026 transition to a stateless architecture, provides the standard interface for agents to connect to live tools and external data sources. A2A handles peer-to-peer coordination, enabling autonomous systems to negotiate tasks and exchange information. Complementary standards — Agent Skills (procedural manifests), ACP (structured communication), and ANP (network-level routing) — fill specific functional niches.

The Systems Layer handles operationalisation: orchestration patterns, registries, gateways, and runtime control loops that allow agents to be deployed as services within enterprise environments.

The Services Layer maps agentic capabilities to classic services-computing concerns — discovery, composition, and economic controllability, including emerging models like the action-based PACT pricing framework.

This is a coherent and increasingly mature architecture. What it does not contain is a governance layer. The distinction matters enormously. Coordination — the technical ability of agents to communicate and collaborate — is not the same as governance — the institutional capacity to make, contest, and audit decisions. The former has been largely solved. The latter has barely been started.

Governance is increasingly viewed as an architectural layer that sits above current interoperability standards — but unlike those standards, it has no home, no working group, and no roadmap.

The Nineteen Gaps

The Sovereign Object OS (SOOS) project, which has conducted one of the most systematic analyses of the current protocol landscape, has identified nineteen distinct technical gaps that prevent agents from safely acting on behalf of humans in consequential domains such as finance, healthcare, and legal services.

Several of these gaps are particularly significant. There is no standard agent intent declaration (Gap G-01): agents operate with implicit, unverifiable scopes, meaning that neither the systems they interact with nor the humans nominally overseeing them can reliably determine what an agent is authorised to do. There is no multi-agent delegation chain standard (G-08): when one agent delegates to another, the chain of authority is not recorded in any verifiable, protocol-level format. There is no standardised human escalation trigger (G-03): agents may fail silently or continue operating indefinitely after authorisation has been withdrawn.

Perhaps most consequentially for regulated industries, there is no machine-readable prohibition standard (G-06) and no jurisdiction-specific regulation encoding (G-07). This forces organisations to rely on bespoke, manual legal engineering for every market in which they deploy agents — an approach that is both unsustainable at scale and systematically difficult to audit.

"The protocols that are becoming critical digital infrastructure were designed to make agents work together. They were not designed to make agents accountable."

The Governance Legitimacy Problem

Beyond the technical gaps, there is a structural governance problem that the technical community has been slower to acknowledge. The AAIF, which now serves as the primary steward of MCP and Agent Skills, is a corporate-backed foundation. Its governance model, while vendor-neutral in aspiration, concentrates standard-setting authority within a small number of large technology companies. As these protocols become critical infrastructure — as consequential to daily life as TCP/IP or HTTPS — the legitimacy of that governance model becomes a serious question.

The contrast with the IETF is instructive. The IETF, which governs the foundational protocols of the internet, operates on a principle of "rough consensus and running code" and maintains a governance structure explicitly devoid of formal government or national delegation control. Its processes are slow and sometimes frustrating, but they produce standards with broad legitimacy precisely because they are open to participation from any qualified contributor. The IETF's agentproto working group, which is now standardising the underlying dialog context for agent interactions, represents a partial move toward this model — but it addresses only a narrow slice of the governance problem.

Research published in October 2026 argues that the current trajectory risks replicating the "security deficit" of early internet protocol development, where foundational design choices made without adequate security consideration took decades to remediate. The analogy is apt. The early internet was designed for connectivity, not security. The early agentic internet is being designed for coordination, not governance. The consequences of that choice will compound over time.

Emerging Responses: SOVP, ATP, and the Sovereign Tech Standards Network

Several initiatives are attempting to address the governance gap, though none yet provides a comprehensive solution.

The protocols that are becoming critical digital infrastructure were designed to make agents work together. They were not designed to make agents accountable.

The Sovereign Validation Protocol (SOVP), an experimental IETF draft published in October 2026, provides a protocol for verifying, before data ingestion, that an identity document was produced by a party controlling a specific DNS-published key. This addresses a narrow but important slice of the trust problem — establishing that an agent's claimed identity is cryptographically bound to a verifiable authority.

The Agent Trust Protocol (ATP) Community Group, proposed at W3C in June 2026, is developing specifications for quantum-safe agent identity (the did:atp method, using a hybrid of Ed25519 and ML-DSA-65 signatures), an agent trust scoring model supported by W3C Verifiable Credentials, and privacy mechanisms including pairwise DIDs, selective disclosure, and zero-knowledge proofs. This is more ambitious — it attempts to build a trust infrastructure that can survive both current and post-quantum threat models.

The W3C DID Methods Working Group, launched in November 2026 and chartered through November 2028, is standardising DID methods that utilise Web technologies, with a focus on ephemeral, web-based, and decentralised web-based methods. Its work on agent-specific use cases — trust graphs, delegation, and behavioural attestation — is directly relevant to the governance gap, though it remains at an early stage.

The Sovereign Tech Standards Network, launched in mid-2026, takes a different approach: providing financial support and mentoring to open-source maintainers, enabling them to contribute directly to IETF, W3C, and ISO working groups. The goal is to bridge the gap between "paper-perfect" standards and real-world implementation experience — ensuring that future protocols are shaped by practitioners who understand the failure modes, not only by theorists who understand the design space.

The Enterprise Governance Stack: A Parallel Architecture

In the absence of standardised governance protocols, enterprises have been building their own. The AI Governance Institute's reference architecture, the ODA3 Institute's GAISSF™ framework, and the Agent Trust Infrastructure (ATI) published in November 2026 all describe variants of a multi-plane governance model that separates policy (what must happen) from decision (real-time enforcement), execution (model and tool operation), and evidence (immutable audit trails).

The Artificial Intelligence Runtime Governance Protocol (AIRGP), published in May 2026, standardises the wire-level exchange of governance signals — policies, verdicts, telemetry — between runtimes, control planes, and audit ledgers, using a JSON-based format over HTTPS/gRPC with tamper-evident logs using SHA-256 hash chains. The Agent Authorization Protocol (AAP) defines how agent identity, delegation, and authorisation are recorded, designed to work alongside MCP and A2A to manage the layer of permission rather than mere connectivity.

These are serious efforts, and they represent genuine progress. But they are enterprise-specific, proprietary in their implementation details, and not interoperable with each other. An agent operating within an AIRGP-governed environment cannot automatically demonstrate its compliance to a system governed by ATI. The governance layer, like the connectivity layer before it, is fragmented — and the fragmentation is becoming a problem as agents increasingly operate across organisational boundaries.

"An agent operating within one governance framework cannot automatically demonstrate its compliance to a system governed by another. The governance layer is fragmented in exactly the way the connectivity layer was before MCP and A2A resolved it."

The Sovereignty Dimension

The governance gap has a geopolitical dimension that the technical community has been even slower to address. Research published in August 2026 identifies a "fragmentation risk" in which the proliferation of disconnected sovereign clouds and agentic AI systems threatens interoperability and economies of scale. The argument is counterintuitive but important: interoperability is a precondition for sovereignty, not a threat to it. Without shared technical protocols, sovereign clouds lose the interoperability and economies of scale that historically characterised cloud computing — and sovereign agent deployments face the same risk.

An agent operating within one governance framework cannot automatically demonstrate its compliance to a system governed by another. The governance layer is fragmented in exactly the way the connectivity layer was before MCP and A2A resolved it.

The IETF's governance model — open, technically meritocratic, devoid of national delegation — has historically provided a neutral ground on which competing national interests could converge on shared technical standards. Whether that model can be extended to the governance layer of the agentic internet, where the stakes are higher and the interests more directly in conflict, remains an open question. NIST in the United States and the ITU internationally are both developing AI agent standards initiatives, signalling a transition from purely industry-led development toward more formal international standards processes. But the timelines for those processes are measured in years, and the deployment of consequential agentic systems is measured in months.

What a Complete Stack Would Require

A genuinely complete agentic protocol stack — one that could support the deployment of autonomous agents in consequential domains with appropriate accountability — would need to address several requirements that current standards do not.

It would need a standardised intent declaration format: a machine-readable specification of what an agent is authorised to do, in what contexts, on whose behalf, and subject to what constraints. This is the foundation of accountability — without it, neither oversight nor audit is possible.

It would need a delegation chain standard: a verifiable record of how authority has been delegated from human principals through orchestrating agents to executing agents, with cryptographic binding at each step. The W3C DID work provides some of the building blocks; the governance layer would need to assemble them into a coherent standard.

It would need human escalation primitives: standardised mechanisms for agents to recognise situations that exceed their authorisation and to transfer control to human authorities in a way that preserves context and audit trail. This is not merely a technical requirement — it is a governance requirement, and it needs to be built into the protocol layer, not bolted on as an application-level feature.

It would need jurisdiction-aware compliance encoding: a way for agents to carry machine-readable representations of the regulatory constraints applicable to their operation, and for the systems they interact with to verify and enforce those constraints. The SOVP draft addresses a narrow version of this problem; a complete solution would need to be far more comprehensive.

And it would need a governance legitimacy framework: a process for developing and maintaining these standards that is open, technically rigorous, and not captured by any single commercial interest. The IETF model provides the closest existing template, but extending it to the governance layer will require deliberate institutional design.

The Window for Getting This Right

The history of internet protocol development offers a cautionary lesson. The foundational protocols of the internet — TCP/IP, DNS, SMTP — were designed for a world of trusted academic and government users. The security and governance problems that resulted from that design choice have cost the global economy hundreds of billions of dollars and remain only partially resolved decades later. The agentic internet is being built now, and the design choices being made now will shape its properties for a generation.

The technical community has demonstrated, with MCP and A2A, that it can move quickly when the problem is well-defined and the incentives are aligned. The governance problem is harder — it requires coordination not just among engineers but among lawyers, regulators, ethicists, and the communities that will be affected by autonomous agent decisions. But the window for building governance into the protocol layer, rather than retrofitting it later, is closing. The AAIF, the IETF, the W3C, and the national standards bodies that are now engaging with this space have an opportunity to get this right. The question is whether the urgency of the governance problem is understood with the same clarity as the urgency of the connectivity problem that preceded it.

The agentic internet is being built. The governance layer that would make it trustworthy is not. That is the most important standards problem of 2026 — and it is not yet being treated as such.

Sources & Further Reading

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
AI agentsinteroperabilityMCPA2A protocolgovernancestandardsIETFtrust infrastructure
The engine behind the Signal

Where this connects to Society OS

The Sovereign Intelligence Hub is the free, open front door of Society OS — the sovereign operating system that turns the ideas you just read into working governance. Where this piece names a problem, Society OS is building the machinery to solve it: AI agents that act with your authority, trust you can verify, and compliance that runs as code.

The 42-Protocol Stack

The governance engine beneath every article — led by the Sovereign Trinity: Human-Twin-Agent identity, HEARTrank trust, and WISE Contracts that execute law, not just code.

F-ACT — the open agent standard

The vendor-neutral framework for governing AI agents before they act: Grant, Usage, Audit, Revocation, Data — free to read, cite and implement.

The Sovereign Platform

Put it to work: govern a fleet of AI agents with verifiable authority, tamper-evident evidence, and compliance-as-code across your whole operation.

Explore membershipRead the F-ACT standard

Continue Reading

More from the Sovereign Intelligence Hub

The Missing Layer in Autonomous Coordination
Standards & Interoperability

The Missing Layer in Autonomous Coordination

11 min read
The missing layer in cross-border autonomy is not intelligence but procedure
Standards & Interoperability

The missing layer in cross-border autonomy is not intelligence but procedure

12 min read
The Protocol Stack That Will Define the Intelligent Economy: MCP, A2A, and the Architecture of Agentic Interoperability
Standards & Interoperability

The Protocol Stack That Will Define the Intelligent Economy: MCP, A2A, and the Architecture of Agentic Interoperability

17 min read
Interoperability is becoming industrial policy by other means
Standards & Interoperability

Interoperability is becoming industrial policy by other means

14 min
How interoperability became a strategic discipline
Standards & Interoperability

How interoperability became a strategic discipline

12 min
How Interoperability Became the Quiet Infrastructure of the Digital Age
Standards & Interoperability

How Interoperability Became the Quiet Infrastructure of the Digital Age

14 min

Never miss a signal

Weekly intelligence, no noise

Governance Toolkit

The Evidence
92 % ungoverned
The Framework
GUARD chain
Your Risk
Sourced model
Self-Assess
No login required

The Sovereign Intelligence Hub — Society OS

© 1989–2026 Society OS Pty Ltd. All rights reserved.