Industries / Critical infrastructure

ICA for critical infrastructure

Govern the agent.
Never the process.

Utilities, pipelines, water systems and nuclear plants hold a hard line between IT and operations. ICA decides what an AI agent may ask for at that boundary, records who authorized it, and proves every halt, starting on IT-side workflows that never command the process.

Request a briefing Verify the proof yourself

What did an AI agent do on 2 October 2026 at 03:09:35 UTC?

Signed ICA receipt · production Checking with ICA

Agent: NVIDIA Nemotron Nano 12B v2 · IN_TRAINING
Asked: DELETE httpbin.org/delete
Class: irreversible
Decision: DENY
Reason: no signed scope contract on record; fail closed
Signed: 2026-10-02 03:09:35 UTC · ed25519

This page asks ICA's public verifier about this exact receipt as it loads. Open the raw check

4 industries, one kind of record

What the reviewer asks for, and what ICA hands them.

Energy and utilities

NERC CIP where bulk electric systems are in scope, OT change control

What the reviewer asks for: Who authorized an action that could touch an OT boundary, and the halt.
What ICA hands them: An authorization and halt record at the IT and OT boundary, on an IT-side workflow first.
Where to start: One IT-side workflow that never touches the bulk electric system.Compliance accepts the boundary: yes or no.
Beside what you run: ICA controls the agent, never the process: it decides what the agent may ask for, and proves it.

Water and pipeline operators

Sector risk management, OT change control, incident reporting

What the reviewer asks for: Who authorized an action at the IT and OT boundary, and the halt.
What ICA hands them: Halt and authorization records at the boundary, verifiable by a regulator after an incident.
Where to start: One IT workflow that never commands the process.Operations accepts the boundary: yes or no.
Beside what you run: ICA sits in the agent's path, upstream of operations, and proves each halt.

Nuclear

NRC cyber security plan

What the reviewer asks for: An authorization record for an agent action inside the plant boundary.
What ICA hands them: Per-procedure authorization records, verifiable offline inside the plant boundary.
Where to start: One procedure, inside the existing boundary.Licensing accepts the record: yes or no.
Beside what you run: ICA records the agent's step inside the procedure the plant already owns.

Water utilities

America's Water Infrastructure Act, state primacy

What the reviewer asks for: A control-action authorization a regulator can see.
What ICA hands them: A halt record per facility, readable by the regulator.
Where to start: One facility.Security accepts the halt record: yes or no.
Beside what you run: ICA governs the agent that asks for remote access, and records who allowed it.

How ICA governs every agent action

Decide. Hold. Halt. Prove.

Decide

Before an agent acts, ICA checks the action against the agent's signed scope. It is allowed, refused, or held for a person. No scope on record means no action: ICA fails closed.

Hold

Irreversible actions wait for a named human. Approvals come only from a signed-in identity, and the rule can require a second, distinct approver.

Halt

ICA halts and quarantines a live agent, and seals the halt to the ledger before the action is stopped.

Prove

Every decision becomes a signed receipt, hash-chained, sealed into a Merkle root and timestamped. Anyone can verify it without asking ALEETH.

Proof, not promises

A real receipt. Check it yourself.

This receipt is real. On 2 October 2026 an AI agent running on NVIDIA Nemotron Nano 12B asked production ICA to delete. It had no signed scope for that, so ICA refused before the call ran. Your browser asked ICA's public verifier about this exact receipt when the page loaded.

Verify a receipt See provenance

Signed ICA receipt · production Checking with ICA

Agent: NVIDIA Nemotron Nano 12B v2 · IN_TRAINING
Asked: DELETE httpbin.org/delete
Class: irreversible
Decision: DENY
Reason: no signed scope contract on record; fail closed
Signed: 2026-10-02 03:09:35 UTC · ed25519
Anchored: root b63347a3211d · leaf 1,820 of 4,251 · 04:29:20 UTC

Receipt fingerprint · SHA-256

9f3f 7c52 9a56 df77

9370 b727 f16e 309b

0ae7 0677 77b5 3c06

847c 6a24 9378 7cc4

Fingerprint matches the receipt bodywaiting
Ed25519 signature valid against ICA's published keywaiting
Included in a signed, anchored Merkle rootwaiting

This page asks ICA's public verifier about this exact receipt as it loads. Open the raw check

AI Governance Readiness Check

Where does your AI governance actually stand?

Answer six questions. Get an indicative Control Readiness Score and the gap between policy and proof, on the spot, with no login. The paid Control Readiness Assessment confirms the score by testing.

What you get

An indicative score from 0 to 100, and the widest gaps between what your AI policy claims and what your controls can prove today. The readiness score is not an ICA certificate.

The score weighs six dimensions of control:

Authority. Is every AI action authorized at the moment it acts, not merely allowed in principle.
Enforcement. Are the limits enforced in the runtime, or only written in a policy.
Evidence. Is there a signed, independently verifiable record of what happened.
Oversight. Can a human intervene, and is that intervention itself recorded.
Provenance. Model, data, and supply-chain lineage behind each decision.
Accountability. Who answers when an autonomous agent acts, and can you prove it.

Indicative, from your profile. The paid Control Readiness Assessment confirms it by testing enforcement and evidence directly.

Other industries

Healthcare Life sciences Financial services Insurance Defense Government Communications Transportation Public companies Education Hospitality

Govern the agent.Never the process.