Energy, utilities and critical infrastructure
Agents for critical infrastructure, safe when things go wrong.
In operational technology a wrong action has physical consequences. MAQPNA keeps agents in isolated sandboxes on your own sites, requires an engineer for safety-relevant calls and fails closed when a control is unavailable.
Why operators keep agents away from operations
Safety, regulation and segmentation come first.
- Physical consequencesSwitching, setpoints and work orders affect equipment and people.
- Segmented networksOT networks are segmented or disconnected; cloud-hosted agents cannot reach them, and should not.
- NIS2 obligationsOperators of essential services must show access control, logging and incident handling.
- Failure modesWhen a dependency fails, the safe default is to stop, not to carry on.
How MAQPNA solves it
Safety-first defaults
Fail-closed defaults
If the revocation list or ledger is unavailable, calls are denied in the sovereign profile.
Engineer approval
Safety-relevant tools need an approver, or two, with separation of duties.
Isolated sandboxes
Each session is isolated, with default-deny egress and tools reachable only through the gateway.
On-site and air-gapped
Install from a signed offline bundle on your own sites, with no phone-home.
Kill switch
Stop an agent, session or user across the installation within about a second.
Incident evidence
A tamper-evident ledger and session timeline for incident reviews and regulators.
Architecture
A maintenance-planning agent on an operator's site
The agent reads asset data and drafts work orders. Anything that touches operations is held for an engineer, and nothing leaves the site.
Sandboxes
tier-1 (microVM)
asset-maintenanceoutage-summaryprocurement-assist
MAQPNA gateway
identity · policy · DLP · budgets · approval · audit
- Allowed
historian.read_seriesread-only-tools - Held for approval
wms.create_work_orderengineer approval - Denied
scada.write_setpointdefault deny
Your systems
- MCPhistorian (read-only)
- MCPwork-management system
- modelon-site model route
Evidence
Evidence for operators of essential services
Controls and evidence, mapped. Not a certification.
- CISA AI in OT principles
the December 2025 joint guidance on integrating AI into operational technology covers AI agents and stresses human oversight and fail-safes. Engineer approval, default deny, fail-closed defaults and isolated sandboxes map to those principles.
- NERC CIP
an agent that touches BES Cyber Systems still falls under access management, security-event logging and change management. The gateway and ledger help produce that evidence for each agent action.
- NIS2 Art. 21 and 23
access control, logging, incident handling and supply-chain evidence for agent actions, with a ledger timeline for the 24-hour and 72-hour reports.
- EU AI Act Art. 12 and 14
record-keeping and human oversight controls. Under Regulation (EU) 2026/1744, Annex III high-risk duties, which include safety components of critical infrastructure, apply from 2 December 2027. How the EU AI Act applies to agents.
- Supply chain
signed releases, SBOMs and an offline install path.
- Honest scope
MAQPNA governs what agents may call; it is not a safety-instrumented system.
FAQ
Common questions
What does MAQPNA do for energy and critical infrastructure?
In operational technology a wrong action has physical consequences. MAQPNA keeps agents in isolated sandboxes on your own sites, requires an engineer for safety-relevant calls and fails closed when a control is unavailable.
What stops energy and critical infrastructure from putting AI agents into production?
Physical consequences: Switching, setpoints and work orders affect equipment and people. Segmented networks: OT networks are segmented or disconnected; cloud-hosted agents cannot reach them, and should not. NIS2 obligations: Operators of essential services must show access control, logging and incident handling. Failure modes: When a dependency fails, the safe default is to stop, not to carry on.
Which controls does MAQPNA enforce?
Fail-closed defaults: If the revocation list or ledger is unavailable, calls are denied in the sovereign profile. Engineer approval: Safety-relevant tools need an approver, or two, with separation of duties. Isolated sandboxes: Each session is isolated, with default-deny egress and tools reachable only through the gateway. On-site and air-gapped: Install from a signed offline bundle on your own sites, with no phone-home. Kill switch: Stop an agent, session or user across the installation within about a second. Incident evidence: A tamper-evident ledger and session timeline for incident reviews and regulators.
Which regulations does it supply evidence for?
CISA AI in OT principles: the December 2025 joint guidance on integrating AI into operational technology covers AI agents and stresses human oversight and fail-safes. Engineer approval, default deny, fail-closed defaults and isolated sandboxes map to those principles. NERC CIP: an agent that touches BES Cyber Systems still falls under access management, security-event logging and change management. The gateway and ledger help produce that evidence for each agent action. NIS2 Art. 21 and 23: access control, logging, incident handling and supply-chain evidence for agent actions, with a ledger timeline for the 24-hour and 72-hour reports. EU AI Act Art. 12 and 14: record-keeping and human oversight controls. Under Regulation (EU) 2026/1744, Annex III high-risk duties, which include safety components of critical infrastructure, apply from 2 December 2027. How the EU AI Act applies to agents. Supply chain: signed releases, SBOMs and an offline install path. Honest scope: MAQPNA governs what agents may call; it is not a safety-instrumented system. MAQPNA supplies controls and evidence; it does not certify compliance.
Where does it run?
The agent reads asset data and drafts work orders. Anything that touches operations is held for an engineer, and nothing leaves the site. The same runtime runs on a laptop, in your cloud account, on Kubernetes or air-gapped.
How do CISA's AI in OT principles apply to AI agents?
The December 2025 joint guidance on integrating AI into operational technology covers AI agents and stresses human oversight and fail-safes. Engineer approval, default deny, fail-closed defaults and isolated sandboxes map to those principles.
Do NERC CIP controls apply to AI agents?
An agent that touches BES Cyber Systems still falls under access management, security-event logging and change management. The gateway and ledger help produce that evidence for each agent action.
Talk to us about a safe first use case.
Policy, approval flow and evidence, walked through with you.
curl -fsSL https://maqpna.com/install.sh | shirm https://maqpna.com/install.ps1 | iex