Skip to content

IAM at the End of the Power Cord

Why your last policy enforcement point is a power outlet

Articles intended for IAM practitioners. You will find here the transposition into a domain where identity and access management is essential and, to this day, almost completely absent.

In North America, a grid operator must remove a departing employee's unescorted physical access to a substation within 24 hours of the termination action. That employee's named user accounts? Thirty days.

That is not an oversight. It is NERC CIP-004-7, R5.1 and R5.3, and it is legally enforceable with penalties that run per day and per violation. The North American regulator has decided that the physical door is thirty times more urgent to close than the login.

Now ask yourself which of the two your IAM program actually governs.

Where Your IAM Chain Physically Ends

Let’s review the chain using the vocabulary we all share: an identity is established and provisioned, an authenticator proves its possession, a PDP evaluates a policy, a PEP enforces the decision along the access path, a session is opened then re-evaluated, a log generates proof, and a review closes the loop.

This chain relies on an assumption we no longer even formulate because it feels so obvious: there is a digital path between the subject and the resource, and we can insert ourselves into it. A reverse proxy, an agent, a SCIM connector, a PAM broker, a domain controller, an application gateway. The PEP is a piece of software sitting on this path.

Remove the digital path, and the entire chain becomes inapplicable. Yet this is the reality for the overwhelming majority of sensitive industrial equipment in service today:

  • A PLC installed in 2003 whose only access control is a key switch on the front panel, with three keys in circulation and a fourth lost six years ago.

  • An electrical cabinet whose opening is protected by a triangular key you can buy online for the price of a coffee.

  • A test bench, a furnace, an autoclave, a press, an EDM machine: equipment where unauthorized use isn't a confidentiality risk, but a risk of physical safety, counterfeiting, sabotage, or misuse.

  • Equipment connected to the site’s power grid, but deliberately disconnected from the IT network—precisely because its operator correctly judged that the connection would create more risk than it would mitigate.

Faced with this asset base, the knee-jerk reaction of the IT world is to demand network connectivity: "connect it, and we will govern it." This is an expensive, slow, and often counterproductive response. Connecting a machine (and especially every machine!)—sometimes designed at a time when network security wasn't part of any specification—just to monitor it, amounts to creating an attack surface... simply to monitor the attack surface you've just created!

There is another way.

Our Thesis: Power Supply Is the Only Universal Enforcement Point

All of this equipment shares one common trait, without exception: they are powered by electricity.

A machine without power does not produce, does not heat up, does not rotate, does not communicate, and cannot be reprogrammed. Power supply is the absolute prerequisite for any use. From an IAM architect's perspective, it possesses remarkable properties:

  • It is universal. No assumptions about protocol, firmware, age, manufacturer, or the presence of an IP stack. The control applies identically to a 1998 PLC and to a machine delivered last month.

  • It precedes the logical layer. A software PEP executes after the target boots and depends on its integrity. An electrical PEP applies before the target exists functionally. No firmware vulnerability on the equipment allows bypassing the absence of power. It is an un-bypassable deny by the resource itself—a rare property we almost never manage to achieve in software architectures.

  • It is binary and stateless. No session to negotiate, no context to maintain in the equipment, no client to install.

The corollary is straightforward: in a sensitive installation, the power outlet is the enforcement point of first and last resort.

This idea is not new. It is simply ungoverned.

Three precedents should reassure any IAM practitioner that we are not proposing something exotic.

Locking out a circuit. Across Europe, EN 50110-1 sets out five safety rules to be performed in order before work begins: disconnect completely; secure against re-connection; verify absence of operating voltage; carry out earthing and short-circuiting; provide protection against adjacent live parts. Germany implements it as DIN VDE 0105-100; the UK adopts it as BS EN 50110-1 on top of the Electricity at Work Regulations 1989.

Its 2023 edition went further, and in a direction that should interest you: it defines a role taxonomy — installation manager, operation controller, work controller, worker — and separates authorization ("formal approval to perform planned work") from permission to start work ("direct instruction to the workers at work location to commence work after all safety measures are taken"). A static entitlement granted upstream, and a dynamic, contextual grant issued on site. That is a PDP and a PEP, described in a electrical safety standard, in 2023.

In the United States, OSHA 29 CFR 1910.147 makes the point even more bluntly. On lockout devices, §(c)(5)(ii)(D) requires that they "shall indicate the identity of the employee applying the device(s)". A padlock that carries a name is a credential. A group lockbox with one personal lock per authorized employee is a multi-holder authorization scheme with non-delegable release — §(e)(3) requires the device be removed by the employee who applied it.

So the industrial world did invent electrical IAM. But an IAM with no directory, no automatic expiry, no queryable log and no remote revocation. It is 1970s IAM, and it is still how most plants run today.

Switched PDUs. In any data centre, per-outlet authorization has existed for a quarter of a century: cut and restore a power port remotely. But the subject of that authorization is an administrative account, never a business identity carrying a qualification. It is an actuator with no policy.

ISO 15118 and Plug & Charge. The most interesting precedent, and the one every IAM practitioner should know. For EV charging, the industry specified X.509 certificate-based authentication of both endpoints, exchanged over the charging cable itself — power-line communication on the pilot line — anchored in a dedicated trust hierarchy: V2G root, contract certificates issued by the mobility operator, charge point certificates, OEM provisioning certificates minted at the factory. The first edition (ISO 15118-2) authenticated the charge point at the transport layer and the vehicle at the application layer; ISO 15118-20 now mandates mutual TLS. A full PKI, with a full certificate lifecycle, whose only purpose is to decide whether electricity should flow, and who gets billed.

The automotive world has already accepted that a power cable is an identity channel. The industrial world is still on padlocks.

Five IAM Reflexes You Need to Unlearn

"physical access control (guards, gates and guns) to protect a control room to restrict access to a group of known personnel to compensate for the technical requirement for personnel to be uniquely identified by the IACS"

— IEC 62443-4-2, definition 3.1.10, compensating countermeasure

The standards committee wrote our thesis for us. Physical access control is what you deploy when the industrial system cannot tell one human being from another. Which is to say: almost always.

This is where applying IT concepts to our domain gets tricky: five principles you would normally apply without a second thought become wrong or dangerous here.

1. "When in doubt, deny" is no longer a universal rule

In IT, fail-closed is the default stance, and trade-offs are almost always decided in favor of security. In OT, denying access can kill. Cutting power to a ventilation system, a cooling unit, a safety loop, or equipment mid-critical cycle is a safety-of-life decision before it is a cybersecurity decision.

Practical consequence: The semantics of denial must be defined equipment by equipment, and this decision does not belong to the security team alone. It belongs jointly to security, operations, and HSE (Health, Safety, Environment). Some equipment will be fail-secure (power cut), others fail-safe (power maintained, logged, alerted), others fail-degraded (reduced functionality allowed). An OT policy model that cannot express these three outcomes is unusable.

This is a nuance that canonical Zero Trust in IT has never had to address.

2. Connectivity is not an assumption

You design your architectures assuming the Identity Provider (IdP) is reachable. In industrial or defense environments, that assumption collapses: sites air-gapped by design, WAN outages lasting days, shielded facilities, contexts where connectivity is actively unwanted.

A PDP that calls a remote service to make a decision is a PDP that refuses to decide when it matters most. The decision must be rendered locally, offline, and remain nominal—not degraded.

You already know this compromise in another form: local verification of a signed token without calling back the issuer for introspection. Same questions, same answers: signed and versioned policies, bounded lifetimes, explicit maximum-freshness revocation lists, trusted time sources, reconciliation upon reconnection. What changes here is the stakes: a stale token accepted by mistake doesn't just grant API access—it grants access to a machine capable of physically hurting someone.

3. Equipment identity cannot be declarative

In IT, we trust what the resource claims about itself because an agent attests to it. In OT, the equipment doesn't know who it is and cannot attest to anything. Worse: it is completely replaceable. Nothing stops someone from unplugging the authorized machine and plugging another one into its place.

Hence an unintuitive but fruitful shift: you do not identify the equipment; you identify the power point and continuously characterize what is connected to it. The electrical and radiofrequency signature of a plugged-in device is an observable attribute—and changes to it become identity events in their own right.

This creates a primitive that does not exist in traditional IAM: rewiring detection as a security signal. A legitimate operator with valid entitlements and a compliant work order who plugs an unexpected piece of equipment into a controlled outlet produces an event that your policy must know how to qualify.

4. The proof must outlive the auditor, not the SIEM

Your logging is designed for detection: high volume, low unit value, short retention, network path to a collector.

Here, it’s the exact opposite. Volume is low: a few dozen events per machine per month. Unit value is extremely high. And the primary use case is almost never real-time detection: it is retrospective proof, examined eighteen months later by an insurer, a prime contractor, a regulator, or an investigating judge to answer one simple question: who powered on this machine on March 14 at 10:40 PM, and on what grounds?

A log sent over the network to a SIEM fails this requirement because the network was likely down on March 14. You need a local, timestamped, chained, signed, non-repudiable log that can be exported independently of any central infrastructure.

5. The subject isn't necessarily human, nor necessarily yours

You are already navigating the wave of non-human identities: service accounts, workloads, proliferating secrets, autonomous agents. The OT world adds a layer IT rarely encounters: a majority of legitimate subjects do not belong to the organization. Temporary workers, tier-2 subcontractors, OEM technicians traveling from another continent, maintenance providers dispatched by a third party.

You must federate an identity, verify an entitlement attribute, and link it to an intervention order—without ever having provisioned the person, often without a directory on the vendor's side, and with no network connection at the time of decision. This is not a solved problem copied from elsewhere; it is an open, fascinating IAM challenge.

What the Regulations Already Say (That Few Have Read)

The idea that IAM belongs to IT while safety belongs to OT does not survive contact with the standards.

What every standards body already agrees on

IEC 62443 is built on seven foundational requirements. The first one, FR 1, is Identification and Authentication Control, and it explicitly distinguishes three classes of subject: human user, software process, and device. The reference security standard for industrial automation puts IAM at the top of its list, and has for years. It is not an IT import — it is a chapter industry never managed to apply to the installed base for lack of a technical means.

The same family goes further than most people realise. IEC 62443-4-2 requires embedded devices to protect their physical diagnostic and test interfaces against unauthorised use (CR/EDR 2.13), and to provide physical tamper resistance and detection (CR/EDR 3.11) — mandatory from security level 2, with logging and notification of tampering attempts from level 3.

It also states the honest objection, and we should state it too: identification and authentication "should not hamper fast, local emergency actions." That constraint is real, and it is exactly why the meaning of "deny" has to be defined machine by machine.

NIST SP 800-82r3, the Guide to Operational Technology Security, places physical security as layer 2 of 5 in its defence-in-depth model — before the network. Its OT tailoring of control PE-3 is unambiguous: physical access controls "are used by the organization when necessary and possible to supplement OT security when electronic mechanisms are unable to fulfill the security requirements of the organization's security plan."

ISO/IEC 27001:2022 restructured Annex A around four themes, one of which is Physical controls (7.1 to 7.14). Two matter here: 7.11 Supporting utilities, the only control in the annex that treats electrical power as a security asset in its own right; and 7.4 Physical security monitoring, new in the 2022 revision — the physical counterpart of logging. The detail sits in ISO/IEC 27002:2022, clause 7.

Where it is already legally binding

Jurisdiction

Instrument

What it demands

North America (electricity)

NERC CIP, enforced by FERC

Authorization of physical and electronic access under one process (CIP-004-7 R4.1); quarterly verification (R4.2); removal of unescorted physical access within 24 hours (R5.1); two or more physical access controls where technically feasible (CIP-006-6 R1.3); alarm within 15 minutes of detection (R1.5); per-individual entry logging (R1.8); 90-day retention (R1.9); escorted visitors with a named point of contact (R2.2); testing of the access control system every 24 months (R3.1)

European Union

NIS2 (2022/2555) and CER (2022/2557)

NIS2 Art. 21(2): access control policies, asset management, HR security, multi-factor authentication, supplier security. CER Art. 13 — the one most commentary misses — requires "adequate physical protection" including access controls, plus employee security management establishing access rights to premises, critical infrastructure and sensitive information by category of critical personnel. Art. 14 adds background checks.

European Union

Cyber Resilience Act (2024/2847)

Annex I requires products with digital elements to be protected from unauthorised access "by appropriate control mechanisms, including but not limited to authentication, identity or access management systems", and to report on possible unauthorised access. It applies to any manufacturer placing a product on the Union market, wherever they are established — a US, Japanese or Taiwanese vendor selling into the EU is in scope and must have an authorised representative or importer carrying the obligations.

Australia

Security of Critical Infrastructure Act 2018 and the CIRMP Rules

The only regime in the world that names both vectors in the rules themselves: a Personnel hazards rule (background checks, periodic re-vetting, plans to minimise unauthorised, unescorted or privileged access to critical components) and a Physical security hazards rule (documented processes to control and limit physical access to critical components, centralised management, continuous monitoring of access controls). A recent tightening moved the regime from principles-based to substantially more prescriptive.

Germany

Critical infrastructure umbrella act (KRITIS-Dachgesetz), implementing CER

Perimeter security, personnel and vehicle access control, staff vetting and training, written resilience plans, 24-hour incident notification. Outcome-based rather than prescriptive: the operator determines what is appropriate and proportionate — and has to document it.

Singapore

Cybersecurity Act and the Code of Practice for CII (CCoP2.0)

Carries an explicit OT addendum, acknowledging that standard IT tooling does not speak Modbus, DNP3 or IEC 61850. Requires detection of lateral movement and credential misuse across both IT and OT, with auditable log retention available on demand for investigation.

What is coming

The United Kingdom is not covered by NIS2. Its Cyber Security and Resilience Bill — still a Bill, in the House of Lords at the time of writing — will amend the NIS Regulations 2018 and pull managed service providers and designated critical suppliers into scope.

In the United States, the picture is less flattering than the European one and worth stating plainly: CIRCIA was enacted in 2022 and its final rule has slipped well past its statutory deadline; the TSA's 2024 proposed rule for surface transport has not been finalised. Outside NERC's electricity sector, the binding American obligations are thinner than the headlines suggest.

Timelines aside, the pressure that actually reaches a plant manager rarely comes from a legislature. It comes down the supply chain, in the security questionnaires and cyber clauses that automotive, aerospace, energy and defence primes are already sending to their tier-2 and tier-3 suppliers.

What they all demand: evidence

These regimes differ in scope, in force and in style. They converge on one thing, and it is the blind spot: they all demand proof. Proof of access control, proof of traceability, proof that qualifications were current. And proof of physical access to an industrial machine exists almost nowhere.

What to Do If You Lead an IAM Program

Inventory critical assets lacking access policies. Not servers, not applications: machines, cabinets, test benches, and equipment where unauthorized use carries physical, safety, or regulatory impact, and for which you cannot produce an access list or audit log. In many industrial organizations, this list easily rivals the number of governed applications—it was simply never compiled because it fell outside everyone's scope (too physical for IT, too digital for facility maintenance).

Define denial semantics equipment by equipment with safety teams. This is the hardest and most critical deliverable. It dictates the whole architecture and cannot be done by the security team alone.

Require an enforcement point capable of offline decision-making. This criterion quickly filters out unsuitable options. Any solution relying on a continuous link to a central service will fail during the first outage—the exact moment risk is highest.

One major don't: Do not connect equipment to the network solely to govern it. Physical access control doesn't require the machine to be smart; it requires the power outlet to be.

Your vocabulary translated

IAM concept

On a sensitive machine

What breaks

Subject identity

The person intervening: operator, in-house maintainer, tier-2 subcontractor, OEM field engineer

Usually not in the operator's directory. Federation with no shared IdP — often no IdP at all on the supplier side.

Resource identity

The machine, the cabinet, the test bench

No cryptographic identifier. IEEE 802.1AR-2018 (IDevID/LDevID) is the right normative answer, but it assumes an identity minted at the factory. The installed base has a nameplate.

Authentication

Proof that the right person is physically present, here, now

No keyboard, no screen, no browser. The factor must be carried by the person and validated by a reader external to the machine.

Authorization

Role and valid qualification and scheduled work order and zone

De facto ABAC. OT specificity: purpose is a first-class attribute. Being qualified is not enough — you need a legitimate reason to be here now. NERC CIP-004-7 R4.1 puts electronic access and unescorted physical access under a single authorization process.

PEP

The actuator that cuts or releases power

Must be physically separate from the machine, or the machine judges its own case.

PDP

The policy engine

Must be local. NIST SP 800-207 describes the PEP as "responsible for enabling, monitoring, and eventually terminating connections between a subject and an enterprise resource" — it assumes the connection exists. Here it does not.

Session

The period the machine is energized

Barely exists: an energized machine stays energized. Needs continuous re-evaluation — the physical counterpart of CAEP / Shared Signals.

Provisioning / JML

Joiner, mover, leaver — but mostly credential expiry

The real OT leaver is a lapsed qualification, not an HR departure. NFPA 70E-2024 §110.4(A)(3): requalification at intervals not exceeding three years. Australia's CIRMP rules: re-vetting of critical workers at least every five years. And note the distinction NFPA 70E draws in §110.4(A)(1)(b) — "a person shall be qualified for certain equipment and tasks to be performed." Least privilege, written into an electrical safety standard.

Access review

Recertifying "may energize machine X"

NERC CIP-004-7 R4.2 already mandates verification of unescorted physical access authorizations once each calendar quarter. Outside that regime, close to nonexistent.

Logging

Timestamped, attributed proof of use

NERC CIP-006-6 R1.8: log each individual, with information identifying them and the date and time of entry. R1.9: retain 90 calendar days. Elsewhere, the only log is the machine's hour counter — unattributed, unsigned, reset at maintenance.

PAM

The OEM engineer's privileged access

PAM covers the remote session. The same engineer, same laptop, same reprogramming capability, standing in front of the cabinet, is covered by nothing.

Revocation

Removing the right to use

NERC CIP-004-7 R5.1: unescorted physical access removed within 24 hours of a termination action — against 30 calendar days for named logical accounts under R5.3. And here, revoking means cutting power, with physical consequences.

Read that table honestly and you will notice you already know how to handle ten rows out of twelve — conceptually. What is missing is not IAM doctrine. It is an enforcement point that acts in the physical world, and a decision point that survives OT conditions.

Expanding the Scope, Not the Stack

IAM spent fifteen years expanding its reach: from employees to customers, customers to partners, humans to workloads, workloads to autonomous agents. At every step, the same objection resurfaced—"that's not really identity"—and at every step, it was proven wrong.

The next frontier belongs to assets that have no IP address and never will. They form the core production engine of European critical infrastructure, are subject to the same regulatory pressures as everything else, and are currently managed with physical keyrings and paper logbooks.

The good news is that your mental model holds up: identity, authentication, policy, enforcement, proof, review. You simply need to accept that the enforcement point is no longer a software component on a network path, but a hardware device on an electrical circuit—and that decisions must be made locally, even when connectivity completely drops.

This reasoning gave birth to Trust-Buster: a local policy engine that verifies the identity and legitimacy of an intervention using physical and radio-frequency environmental signals, actuators that enforce decisions on the equipment's power supply, and audit-ready proof logs—operating seamlessly, connected or offline.

Electrical IAM, in short. The exact same discipline you practice, just with a circuit breaker at the end of the chain.