Proposed open standard  ·  Practitioner-led  ·  v1.0 draft

Every major infrastructure shift has demanded a new operating model.

Infrastructure needs an authority layer.

AI now writes and executes infrastructure changes faster than any team can review them — and it can only be trusted inside infrastructure it actually understands.

The Infrastructure Operating Model is that layer: the authoritative source of what exists, what is intended, and what is legitimate — established before anything acts. Security, automation, compliance, and AI are all downstream of it.

Five questions infrastructure must answer — in real time
  • What exists right now, across all environments?
  • What is this environment intended to do?
  • What constraints and boundaries apply?
  • Who owns the blast radius of this action?
  • Is this action legitimate in this context?
IOM provides authoritative answers to all five.
What an IOM is

An Infrastructure Operating Model is the authoritative layer that defines what infrastructure exists, what it is intended to do, and what actions are legitimatebefore execution occurs.

What an IOM changes

Concrete shifts — not abstractions.

The same operations, run with authority instead of inference. Each row is a cost you carry today, and the capability that replaces it.

Before IOMAfter IOM
Reactive securityPre-execution validation
Documentation driftLiving documentation
Manual approvalsAuthority-driven automation
Slow repatriationModel-based repatriation
AI requires human reviewGoverned AI execution
The category

Everything is downstream of the authority layer

Establish one layer that holds what exists, what is intended, and what is legitimate — and that answers to it before anything executes — and the rest of the operating model follows. These are not separate problems. They are the same problem, seen from different disciplines:

SOURCES OF LEGITIMACY Intent Policy Ownership Relationships Constraints THE AUTHORITY LAYER Infrastructure Operating Model The source of legitimacy for everything that acts on infrastructure. Security AI Automation Compliance Operations GOVERNED DISCIPLINES all of which act on — EXECUTION ENVIRONMENT Cloud · Network · Identity · Applications · On-Prem · Edge

Solve the authority layer once, and every discipline inherits the answer — documentation, repatriation, and platform engineering included.

Why now

Five eras of execution. None added authority.

Every era of infrastructure made execution faster, broader, and more autonomous — and none established authority over what executes, because a human was always in the loop to supply it. The fifth era removes the human.

Era 1
Physical Infrastructure
Direct control, configured by hand
Era 2
Virtualization
Abstracted from hardware; density
Era 3
Cloud
On-demand, elastic, API-driven
Era 4
Infrastructure as Code
Versioned, repeatable, reviewable
Era 5 · now
AI-Operated Infrastructure
Systems that act at machine speed
Therefore
The Authority Layer
Decides what is legitimate, before execution

Infrastructure built for human decision speed cannot govern infrastructure operating at machine speed.

Therefore authority becomes mandatory.
Not a faster tool — a layer that decides what is legitimate before execution.

See the full five-era argument →

The cost of inaction

The authority gap

When no layer holds authoritative understanding, every other discipline pays the interest.

Executives buy pain. Architects buy architecture. The same gap that an architect sees as a missing layer, a CIO experiences as escalating risk, unexplained spend, and audits that never end. The cost of operating without authority does not stay flat — it compounds every year the gap stays open.

Without an Infrastructure Operating Model

  • AI cannot be trusted to act. Agents infer intent from telemetry instead of reading it, so autonomous action stays gated behind human review.
  • Security becomes reactive. Exposure is discovered after it exists, not prevented before it is created.
  • Documentation diverges from reality. What is written down drifts from what is running within hours of every change.
  • Cloud spend becomes difficult to explain. Cost is reconstructed from bills after the fact rather than attributed to owners and intent.
  • Compliance becomes a project. Evidence is assembled by hand for each audit instead of produced continuously.
  • Outages become investigations. When something breaks, ownership and blast radius are reconstructed under pressure rather than known in advance.
The burning platform

Why infrastructure fails

The way infrastructure is understood has not kept pace with how fast it now changes. Five failures surface again and again — and underneath them all is a single root cause.

01

Ownership is unclear

Nobody can prove accountability.

02

Documentation is stale

Reality diverges from records.

03

Automation lacks context

Scripts execute without understanding.

04

Security reacts after change

Controls verify after execution.

05

AI has no authority

AI can reason — but cannot determine legitimacy.

No authoritative layer holds the truth of the environment.
That is the wall — and it is why this is a now problem, not a someday one.

The old way

Infrastructure is still run from the gaps between tools.

Every team already has discovery, infrastructure-as-code, ticketing, monitoring, and a wiki. None of them holds the authoritative truth — so the real operating model lives in the seams between them: disconnected tools, stale documents, tickets, and tribal knowledge. That is not a tooling problem. It is the absence of an operating model.

What we’re ending

Operating from the seams

  • Tools, documents, and tickets each holding a partial, unreconciled truth
  • Records that drift from reality the moment they’re written
  • Governance that verifies after the change has already happened
  • Ownership and blast radius reconstructed during incidents
  • AI and automation acting on inference, not authority
What replaces it

One authoritative layer

  • A single layer holding what exists, what is intended, and what is legitimate
  • Intent made explicit, versioned, and machine-readable
  • Governance that runs before execution, not after
  • Ownership and blast radius known before any change
  • AI and automation bounded by authoritative context

The enemy was never the tools.
It is operating without a layer that makes them coherent.

What is IOM

A precise definition

"An Infrastructure Operating Model is a governance framework that establishes authoritative, continuously reconciled understanding of infrastructure state, intent, and ownership — and uses that understanding to validate actions before execution."

Its purpose is not automation.
Its purpose is authority.

Most infrastructure environments today consist of execution systems that act, control systems that enforce isolated rules, and documentation that lags reality. What is missing is a layer that understands the system as it exists now, knows what is allowed and why, and can determine legitimacy, impact, and accountability before actions occur.

IOM does not slow execution. It makes speed safe. By establishing authoritative understanding as a precondition for action, organizations can move faster with confidence — not slower with caution.

Where IOM sits

INTENT — THE MODEL What exists What is intended What is legitimate EVERY PROPOSED ACTION Human change CI/CD pipeline AI agent THE AUTHORITY LAYER checks each action against intent — before it executes legitimacy ownership blast radius ADMITTED legitimate — passes to execution BLOCKED not what intent allows EXECUTION Cloud Network Identity On-prem

Nothing reaches execution without passing through the layer.

IOM does not replace tools — it governs how they are used.

Why not before · Why always

If it’s essential, why didn’t it exist before?

The function an IOM performs is not new. For as long as infrastructure has existed, something has decided whether a change was legitimate before it happened. For decades that something was people — which is exactly why it was never built as a system, and exactly why it has to be one now.

Until now

Humans were the authority layer

What was allowed lived in senior engineers’ heads, and change moved at the speed people could review it — runbooks, change advisory boards, ticket approvals. The precursors each held a fragment and let the rest drift: the CMDB was a record, not a gate; ITIL change boards governed, but by hand; infrastructure-as-code encoded intent, but only at apply time, one tool at a time. It held together because people were the bottleneck.

What changed

The actor is no longer the one who understands

AI and automation removed both the pacing and the context. An agent acts at machine speed and carries none of the tacit “what’s allowed” that used to live in the people reviewing the change. Take away the human bottleneck and the informal authority layer doesn’t get faster — it disappears. The need didn’t arrive with AI; AI made it impossible to keep meeting that need informally.

From here on

The need is structural, not a trend

Wherever the things that change a system differ from the parties who define what is legitimate, an externalized, enforceable authority is required to bridge them. New clouds, models, tools, and workflows don’t close that gap — they add actors and widen it. Even a single integrated vendor stack still changes over time and still has to reconcile what happened against what was intended. Integration removes interoperability friction; it does not remove the need for authority.

IOM is not a new burden. It is the formalization of a function that always existed —
now that the thing acting is no longer the thing that understands.

Why now

AI, automation, and security break without it

IOM is not important in the abstract. The three forces reshaping infrastructure today each fail the same way — they act without authoritative understanding of what they are acting on. That is the gap IOM closes, and it is why this is a now problem, not a someday one.

AI

AI Requires Authority

Without authoritative infrastructure understanding, AI does not automate work — it automates uncertainty. Agents need to know what exists, what is intended, and what is legitimate before they act.

SEC

Security Requires Intent

Zero Trust verifies identity. It cannot verify legitimacy — whether an action is what the architecture intended. IOM supplies the intent layer that turns “who is this” into “should this happen at all.”

AUT

Automation Requires Admissibility

Execution without validation scales mistakes at machine speed. Pipelines and agents need a pre-execution check that a change is admissible against intent — not detection after it has already run.

Convergence

Five unrelated pressures are converging on the same answer.

No single trend is driving this. Five forces — with different owners, different budgets, and different timelines — are independently pushing organizations toward an authority layer.

AI governance

Autonomous agents need explicit authority before they act — not human review after. Governance has to move upstream of execution.

Repatriation

Moving workloads off cloud requires an authoritative model of what exists and what it is allowed to do — or the move is a rediscovery project.

Platform engineering

Golden paths only hold if intent is enforceable, not just templated. Platforms need a model that validates what teams build against what is permitted.

Zero Trust

Identity is verified; the legitimacy of the action is not. That unverified half — whether a change is what the architecture intended — is the authority layer.

Regulatory compliance

Auditors increasingly want provable intent and pre-execution control, not after-the-fact logs reconstructed under deadline.

Five different owners. Five different budgets.
One missing layer underneath all of them.

Security · The missing layer

Zero Trust secures execution.
IOM governs admissibility.

Zero Trust answers one question well: is this actor allowed to act? IOM answers the question that comes first — should this action happen at all, and is it what the architecture intended? Identity is necessary. It is not sufficient.

A perfectly authenticated actor can still execute an illegitimate change: a valid credential making a change no blueprint authorized, into an environment it was never meant to touch. Zero Trust verifies the who. IOM verifies the whether.

Zero Trust verifies identity.
IOM verifies legitimacy.

What admissibility adds, upstream of execution

  • Intent, not just identity. A change is checked against what the architecture declared — not only against who requested it.
  • Legitimacy before access. The action is evaluated for whether it should occur, before credentials ever reach the resource.
  • Blast radius as a control. The downstream impact is known and bounded before execution — not investigated after a breach.
  • Evidence by default. Every admitted and rejected action is recorded against the intent that authorized it.

IOM does not replace Zero Trust — it sits above it. Identity proves who is acting; admissibility proves the action is legitimate. Together they secure both the actor and the act.

The forcing function

AI requires governed infrastructure

Traditional automation executes instructions. AI creates them. That single change moves the burden of safety upstream — from reviewing what was done to authorizing what may be done.

An agent that writes its own actions cannot be governed by approving its output after the fact. It can only be governed by a layer that defines, in advance, what it is permitted to do and why.

Without authority, AI scales uncertainty.
With authority, AI scales execution.

Before autonomous systems can act safely

  • Ownership must be known. Every element needs a machine-readable owner before an agent touches it.
  • Intent must be defined. The system must read what an environment is for — not infer it from logs and metrics.
  • Blast radius must be understood. The downstream impact of an action must be deterministic before it executes.
  • Accountability must be established. Every autonomous decision must trace back to the intent that authorized it.

IOM provides the authority substrate that AI requires — the authoritative context and explicit boundaries that turn autonomy from a risk into an asset.

The question is no longer whether AI will operate infrastructure.
It is whether infrastructure will have an authority layer before AI does.

The economic story

What happens without an IOM

The difference is not tooling — it is where the cost and the risk live. Without an operating model, every discipline pays after the fact. With one, authority moves before execution.

Without an IOMWith an IOM
Manual auditsContinuous evidence
Reactive securityProactive validation
Static documentationLiving documentation
Slow onboardingBlueprint deployment
Unknown blast radiusPredictive impact analysis
AI restricted to copilotsGoverned AI execution

Each row is the same move: authority established before execution, not effort spent after it.

What an IOM saves

Architects buy the standard. CIOs buy the economics.

Authority established before execution removes cost that is otherwise paid after it. We don’t publish a single ROI figure — the inputs are yours. Here is the lever, the mechanism, and the number you already track.

What it reducesMechanismMeasure it by
Audit preparationContinuous, queryable evidence instead of manual collectionHours per audit cycle
Change failuresPre-execution validation against intent and blast radiusChange-failure rate & rollbacks
Migration timeA living model of what exists and what depends on itTime-to-cutover
Repatriation costModel-based moves instead of per-workload rediscoveryEffort per workload moved
AI enablementAgents act within authoritative, enforceable boundariesShare of actions safe to automate
Documentation effortDocumentation generated from the model, not maintained by handHours maintaining docs

Model your own numbers →

Understanding
precedes
execution.

Every automated or AI-driven action must be validated against authoritative understanding before it occurs. This is not a gate. It is a foundation.

When infrastructure operates without a governing model, speed and control trade off against each other. IOM resolves this trade-off structurally — making every action faster to approve because the basis for approval already exists.

Why this feels familiar

The SD-WAN moment for infrastructure governance

The industry has already lived this shift once — at the network layer. The same move is now arriving for the entire infrastructure stack.

Before

Policy lived device by device

Network policy was implemented box by box. Intent was re-expressed in every device's configuration, drift was inevitable, and no single place held the authoritative picture of what the network was supposed to do.

After SD-WAN

Intent defined once, enforced everywhere

SD-WAN introduced a centralized model: intent is declared once and enforced consistently across the fabric. The network stopped being a collection of independently configured devices and became a governed system.

The Infrastructure Operating Model applies the same principle across the entire stack. Intent is defined once. Understanding is maintained continuously. Actions are validated before execution.
Category definition

What IOM is not

As interest in IOM grows, adjacent categories will claim equivalence. These distinctions are architectural — not matters of degree.

Not

A digital twin

Digital twins describe infrastructure through observation and simulation. They are advisory — positioned outside the execution path and anchored in runtime state. An IOM governs what infrastructure is permitted to exist and do. Its source of truth is declared intent, not observed reality. The distinction is architectural: digital twins observe; IOMs authorize.

Not

A CMDB

CMDBs record assets and configurations in a human-maintained, periodically updated snapshot. They are systems of record — accurate at the time of entry, stale within hours of update. An IOM is continuously reconciled against live infrastructure state, encodes intent and constraints, and makes governance decisions. It is operational, not archival.

Not

A control plane

Control planes execute. They manage infrastructure state by issuing commands to execution systems. An IOM determines what is legitimate to execute — before execution occurs. It is upstream of the control plane, not an alternative to it. IOM decides; the control plane acts on those decisions.

Not

A policy engine

Policy engines evaluate rules. An IOM evaluates rules in the context of an authoritative, continuously reconciled model of infrastructure state, intent, ownership, and relationships. Rules without architectural context cannot determine legitimacy, blast radius, or accountability. Context is not a feature — it is the foundation.

How IOM works

Six core components

A functioning IOM requires all six components. Systems that lack any one of them are incomplete — they may provide insight or enforcement in isolation, but they do not constitute a governing operating model.

01

Canonical state and relationship model

A continuously updated model of infrastructure elements — cloud, network, identity, services — including relationships, dependencies, and environmental context. Not a snapshot: a living representation reconciled against reality in real time.

02

Intent and constraints

Explicit, machine-readable definition of what outcomes an environment is designed to support, what configurations are permitted, and what architectural, regulatory, and risk constraints apply. Intent is never inferred — it is declared.

03

Blueprints

Reusable governance references encoding approved patterns for environment design, network segmentation, security posture, and compliance boundaries. Blueprints are not IaC — they are portable architectural intent used to validate execution paths across environments and platforms.

04

Continuous reconciliation

Continuous comparison of intended state, built state, and actual running state. Reconciliation enables drift prevention rather than drift detection — catching divergence before it becomes exposure, not after it becomes an incident.

05

Contextual validation and authority

Before execution, every action is evaluated against the canonical model: Is this allowed in this context? What is the blast radius? Who is accountable? Only validated actions proceed. This is the defining capability — governance that operates before execution, not after impact.

06

Evidence and learning capture

Every validated decision generates traceability from intent to action, producing audit-ready evidence as a natural byproduct of operation — not as a retrospective documentation effort. Over time, decisions refine blueprints and intent, creating compounding governance intelligence.

The IOM operating loop

Execution never bypasses understanding. The loop is continuous — every execution cycle feeds the model, and the model governs the next execution cycle.

Observe
State + telemetry
Model
Canonical state
Validate
Context + authority
Execute
Governed action
Learn
Evidence + refinement
Understanding governs execution.
Outcomes

What changes when IOM is in place

Without IOM
Probabilistic change. Changes are validated by human review and past experience. Whether a change is safe depends on who reviews it and what they happen to know.
Post-hoc cost attribution. Cloud costs are assembled from provider bills after the fact, requiring manual tagging and reconciliation to trace to owners and purposes.
Episodic compliance. Compliance posture is assessed periodically through audits, point-in-time reviews, and manual evidence collection. Status is always partially stale.
Opaque AI autonomy. AI systems infer intent from telemetry, producing recommendations that require human validation. Autonomous action creates unacceptable risk.
Human escalation as default. Engineers are continuously pulled into decisions that automation cannot resolve because authority is undefined.
With IOM
Predictable change. Every change is evaluated against authoritative intent and constraints before execution. Conforming changes auto-approve. Violations are blocked deterministically. The outcome is known before execution begins.
Continuous cost explainability. Ownership and intent are encoded in the model. Cost attribution is a query, not an investigation — because the model knows what everything is and who owns it.
Continuous compliance. Compliance posture is derived from the continuously reconciled model, producing real-time evidence as a byproduct of operation. Audits become queries, not projects.
Governed AI autonomy. AI systems consume authoritative infrastructure context, explicit intent boundaries, and ownership semantics. Autonomous action is bounded, explainable, and auditable by design.
Accountability without escalation. Authority is encoded in the model. Most decisions resolve automatically. Human escalation becomes an exception path — reserved for genuinely novel situations — not the default.
Buyer's guide

Evaluate any IOM solution against the standard

A neutral, capability-based framework for assessing any solution that claims IOM capability — the criteria to require, the questions to ask, and the claims that don't qualify. No products named, no rankings.

Evaluate by capability

Demonstrated, not described

Ask any solution to demonstrate each mandatory capability against your own infrastructure. If any is absent, it is not a conformant IOM.

The standard

A versioned, vendor-neutral specification

The IOM Standard uses RFC 2119 normative language (MUST, MUST NOT, SHOULD, MAY) to define unambiguous conformance criteria. It specifies what an IOM must provide, what it explicitly does not, and how implementations are evaluated — against the published standard, not a vendor's interpretation.

v1.0 Draft · Foundational Release

Seven mandatory requirements

Authoritative state modeling · explicit intent & blueprint governance · continuous reconciliation · pre-execution validation · explicit ownership & blast-radius awareness · automated evidence · AI-safe reasoning substrate.

See conformance requirements
Governance

Practitioner-led. Vendor-neutral.

The IOM Standard is not owned by any vendor. It is governed as a community resource — developed by practitioners, for practitioners. No company's product roadmap determines the direction of the standard.

Standard stewardship

Open governance

The standard is maintained as part of the IOM standards initiative. Changes are proposed and refined through open review, and versioned with public change memos.

Who can participate

Practitioners and organizations

Enterprise practitioners, architects, security leaders, and technology organizations are invited to contribute to working group deliberations, propose standard refinements, and endorse the standard as co-signatories.

Conformance

Conformance program (in development)

A conformance test suite is in development. Implementations — commercial or open-source — will be evaluated against the published standard requirements, not against any reference implementation.

Get involved

Three ways to participate

The IOM Standard grows stronger with practitioner input. Whether you want to apply IOM in your organization, contribute to the standard's evolution, or endorse it as an industry initiative, there is a place for you.

Apply

Get the IOM starter kit

A practical guide to evaluating your organization's infrastructure governance maturity, mapping existing tools against IOM requirements, and planning a structured adoption path.

View the starter kit
Contribute

Join the working group

Working group members contribute to standard refinements, conformance criteria development, and implementation guidance. Membership is open to qualified practitioners and organizations.

Apply for membership
Endorse

Sign the standard

Organizations and practitioners who have reviewed and validated the IOM Standard can endorse it as co-signatories — adding institutional credibility to the community initiative.

Become a co-signatory