Utility Data Contracts for Agentic AI: CIM, IEC 61850, and OT Boundaries

Agentic AI pilots do not fail only because the model is weak. They often fail because the workflow has no data contract.

In a utility, “data” can mean CIM-aligned network records, IEC 61850 substation context, CMMS/EAM work orders, historian exports, SCADA/EMS context, DGA files, inspection notes, relay events, spreadsheets, PDFs, and expert assumptions. If an AI agent can touch those sources without a contract, the pilot becomes hard to govern.

Use the mapper above as a client-only planning aid. It does not upload data, certify standards compliance, create a data-migration plan, connect to SCADA, or approve operational decisions. Tool inputs stay local in the browser; optional site analytics, when accepted, does not receive selected tool values.

Why agents need a contract

Open-ended AI access is a poor fit for critical infrastructure. A utility agent needs to know what it is allowed to see, what each source means, who owns the evidence, what should be redacted, and where a draft must stop for human review.

IEC 61970 and IEC 61968 are useful public anchors because they define integration and information-model concepts used across utility systems. IEC 61850 is useful because it structures power utility automation and substation communication context. Those standards do not make an AI workflow safe by themselves, but they help teams ask better questions about identity, mapping, provenance, and boundaries.

NIST SP 800-82 Rev. 3 and the NIST AI RMF add the governance lens: OT security, access control, risk mapping, and accountable system behavior must be designed before broad automation.

The utility data-contract checklist

Contract layer	Question to settle	Why it matters	GridAPM pilot artifact
Asset identity	Which transformer, feeder, substation, bay, winding, bushing, or work queue is in scope?	Agents cannot organize evidence if asset identity changes between systems.	Safe asset identity map with approved aliases.
Source ownership	Who owns each source, export, file, model, or record class?	Reviewers need a responsible owner when evidence is stale, missing, or conflicting.	Source register with owner and sensitivity.
Field semantics	Which fields, units, timestamps, quality flags, and versions are required?	AI drafts become unreliable when units and timestamps are ambiguous.	Evidence dictionary and normalization notes.
OT boundary	Which systems are read-only, excluded, redacted, or never connected?	Planning assistance must be separated from live operations and control.	Data boundary and no-control statement.
Review authority	Who can approve, reject, edit, or escalate an AI draft?	Agentic AI should not become shadow authority.	Reviewer path and audit trail.

What the agent can do after the contract exists

Once the data contract is clear, a bounded agent can help with practical evidence work:

• Summarize approved transformer records.
• Flag missing source ownership, units, dates, or versions.
• Draft reviewer questions.
• Identify conflicting records.
• Prepare a human-reviewed evidence pack.
• Record rejected draft patterns for closeout learning.

That is very different from giving an agent direct operational authority.

What GridAPM should not claim

Public copy should stay strict. A data-contract pilot should not imply:

• IEC or CIM certification.
• Plug-and-play integration with all utility systems.
• SCADA, EMS, DERMS, or protection-control authority.
• Data migration, master data management, or cybersecurity certification.
• Final diagnostic conclusions.
• Automated maintenance or operating decisions.

The stronger and more credible claim is that GridAPM helps utilities structure approved evidence for a controlled, local-first, human-reviewed pilot.

A practical pilot pattern

Step	Utility activity	Agentic AI assist	Human review point
1. Source inventory	List approved systems, exports, owners, sensitivity, and exclusions.	None, or only template guidance.	Data and security owners approve scope.
2. Contract map	Map identity, fields, units, timestamps, versions, and source quality.	Flag gaps and draft mapping questions.	Integration and asset teams validate meaning.
3. Draft package	Prepare evidence summaries and gap lists from approved records.	Draft source-linked review packages.	Qualified reviewers approve or reject draft claims.
4. Evidence pack	Package reviewed evidence, assumptions, gaps, and decisions.	Format the package and preserve traceability.	Workflow owner signs off on pilot results.

How GridAPM helps

GridAPM can help utilities, TSOs, DSOs, generation teams, and industrial energy teams evaluate a controlled data-contract pilot by focusing on transformer APM and evidence workflows first.

Useful internal links:

• Integrations for DGA, PRPD, SFRA, CMMS/EAM, historian, SCADA context, and pilot data paths.
• Data handling for public positioning around evidence, retention, and review.
• Security and trust for local-first and human-reviewed boundaries.
• Pilot evaluation and sample evidence pack for buyer-ready next steps.
• Tools hub for more client-only planning tools.

The data-contract principle

Do not ask an agent to understand the utility before the utility has defined the contract.

For GridAPM, the first win is not broad automation. It is a source-linked, human-reviewed transformer evidence workflow where data boundaries are explicit enough for engineers to trust the review process.