Industries · Utilities & Energy · Governance Substrate
Your operations are adding AI agents near the most constrained networks you run.
Electric and energy utilities are putting AI agents into operations and the work that sits next to operational technology. Those networks are the most security-constrained in your enterprise, and the accountability question arrives before any regulator writes the rule. When something goes wrong, you will be asked what the agents did, on whose authority, and whether you can prove it.
§ 01 The problem
The obligation arrives before the rulebook.
No standard yet specifically governs autonomous agents in utility operations. NERC Project 2026-02 treats AI as a load, not as an operator. The obligation to prove what an agent did arrives before the rule that names it, and it arrives the first time an agent touches a system that matters.
The moment an agent acts on a real operational system, the question is no longer model quality. It is control: which agent acted, on what system, under which policy, and whether the record survives a reliability review. Most AI programs cannot answer that. Context is lost across sessions, handoffs leave no trail, and the model behind the agent can be throttled or revoked by forces outside your control.
That is not a model problem. It is a substrate problem. It is the layer Subnet345 builds.
§ 02 Inside the OT perimeter
Governed where it happens. Audited there.
Inside your OT perimeter: every agent action is governed by policy you write, attributed to a specific agent under that policy, and hash-chained into a tamper-evident audit chain that ships to retention inside your boundary. The substrate runs where you run, on hardware you control, with no outbound dependency at the governance and audit layer.
§ 03 The air-gap question, answered honestly
Air-gap-capable by construction.
OT networks are segmented and isolated, and NIST SP 800-82r3 is candid that a true air-gap is rarely found in practice. We do not assume your network is air-gapped. We build so that it does not have to be reachable from outside to be governed.
Today, the governance, audit, attribution, and policy substrate deploys air-gap-capable inside your OT perimeter, with no outbound traffic at that layer. The agent inference layer still reaches frontier model APIs from the operator-side network. Sovereign-model deployment inside the perimeter, where even the model behind the agent runs without outbound traffic, is on the roadmap through our model-training pipeline.
The substrate is air-gap-capable today; the models are not yet. We say which is which, because an operator who runs a control room will check.
§ 04 Frameworks
Mapped to the rules you already answer to.
No framework yet prescribes controls for autonomous agents in utility operations. These are the obligations the work sits inside today, and what the substrate produces against each. This material is not legal or regulatory advice; assess your specific obligations with qualified counsel.
NERC CIP
Mandatory cybersecurity standards for the Bulk Electric System (CIP-002 through CIP-014). Agents that touch in-scope systems inherit those control obligations; the substrate produces the attributable, tamper-evident record those controls assume.
FERC
Approves and enforces the NERC reliability standards, including CIP. When an AI-influenced decision has to survive a reliability or prudence review, the trace is the difference between an answer and an argument.
NIST AI RMF 1.0
The AI Risk Management Framework and its Generative AI Profile (NIST AI 600-1) ask for governed, mapped, measured, and managed AI. Attribution and an interrogable audit record are how you evidence the Manage function for agents.
NIST SP 800-82r3
Guidance on securing operational-technology environments. It is candid that true air-gaps are rarely found in practice; OT is segmented and isolated. The substrate is built to run inside that boundary without reaching out of it.
The question we help you answer
Can you prove what your agents did near the systems that cannot fail?
If you cannot answer that today, and a reliability review asks it tomorrow, the substrate is what changes the answer. We are accepting design partners in Utilities & Energy now.