ABOUT — WHY WE EXIST

AI is being handed the keys to systems where failure is not an option.

Someone has to stand in the path and rule on what crosses. We build the deterministic enforcement layer — and publish exactly what each product can evidence today.

The containment.ai team
PLATE P-05 — THE TEAM
THE WAGER WE'RE MAKING
AI is crossing into systems where a wrong action costs more than a wrong answer — weapons programs, power grids, actuators, export-controlled data rooms. The industry's default safeguard is another model watching the first one: probabilistic, unexplainable, untestable at the moment it matters.
Our wager is older than AI: put a deterministic control at the boundary, make it fail closed, and record each ruling for review. If the exact input, context, and policy bundle are retained, the policy decision can be re-evaluated. If we're right, the perimeter — not the model — is where trust in AI gets built. We run daily operations under the workflow controls now; the internal action guard remains shadow/observe until the evidence supports enforcement.
WHERE WE SIT — TWO LAYERS
The control plane is the incumbents' layer — Microsoft calls Agent 365 "the control plane for AI agents"; ServiceNow, Cisco, Check Point, and Palo Alto sell the same altitude. It decides which agents exist and watches what they do: discovery, identity, observability, kill switches. Keep it — we don't compete for that line item.
Containment is the enforcement layer beneath it: deterministic permit, clamp, deny, or quarantine at the seam where an agent's tool call, an OT command, or a robotics actuation becomes irreversible. Deterministic policy—not an AI model—makes the enforcement decision, and each governed ruling leaves a product-specific decision record. They claim breadth across the enterprise; we claim depth at one seam. A control plane can sit on top of us — most should.
WHAT WE ARE — STATED FOR THE RECORD
Containment.ai is the AI Action Enforcement Layer — the independent enforcement layer for AI-powered systems.

For organizations putting AI to work where actions have consequences, Containment.ai rules on what AI is allowed to do before it does it: deterministic policy evaluation at the point of action — permit, clamp, deny, or quarantine. Each governed ruling leaves a product-specific decision record.

In plain English: Containment.ai puts a policy decision between what AI wants to do and what your systems allow.

THE PROMISE
No consequential AI action crosses a governed boundary without a policy ruling and a record.
IF YOU HAVE 30 SECONDS
AI stopped just answering questions. It acts — it calls tools, moves data, commands machines. When a wrong action costs more than a wrong answer, somebody has to rule on what executes. Containment.ai is the AI Action Enforcement Layer: we intercept the consequential action before it runs, use versioned policy—not an AI model—to make the enforcement decision, enforce the ruling in-path, and leave a decision record for review. Keep the control plane. Add authority at the action seam.
IF YOU HAVE TWO MINUTES
The shift. AI changed from answering to acting: prompts became tool calls, OT commands, and actuations — into weapons programs, power grids, and export-controlled data rooms, where a wrong action costs more than a wrong answer. The gap. Your existing stack — inventory, identity, observability — tells you what exists, who acted, and what happened. Necessary, and insufficient: none of it is an authorization decision at the moment of action. The layer. Containment.ai sits beneath the control plane at the seam where an action becomes irreversible. One discipline on every surface — intercept, canonicalize, evaluate, enforce, record — deterministic and fail-closed; no AI model makes the enforcement decision. The proof. Every governed ruling leaves a product-specific decision record; Gateway staging receipts are Ed25519-signed and hash-chained, while connected-product signing and verification methods are stated on Trust. We run our own company on the product — ten agents, failures published — and our Trust page is the canonical record, outranking our marketing. The fit. A connected tier for everyday enterprise AI, and the Mission Authorization Gateway — our flagship — for mission boundaries in national security, defense tech, and aerospace & defense.
COMPANY BOILERPLATE — CANONICAL, PRESS-READY

Containment.ai builds the AI Action Enforcement Layer — the independent enforcement layer for AI-powered systems. It delivers deterministic rulings on consequential AI actions at the moment of execution. No AI model makes the enforcement decision, and each governed ruling leaves a product-specific decision record. The platform governs employee AI use in the browser, autonomous agent actions, and mission boundaries through one discipline — intercept, canonicalize, evaluate, enforce, record — built for national security, defense tech, and aerospace & defense, and for any operation where a wrong action costs more than a wrong answer. Containment.ai is headquartered in Great Falls, Virginia. The canonical record of what we run and what we claim is public at containment.ai/trust.

HOW WE OPERATE
PRINCIPLE 01
Deterministic policy holds the gavel—not an AI model
The enforcement decision is deterministic policy evaluation. That's not a feature choice; it's the company.
PRINCIPLE 02
Claim only what's attested
We claim only what's attested. Designed-against means designed-against. Our Trust page is the canonical record, and it outranks our marketing.
PRINCIPLE 03
Run on our own gateway
Our prompts and agents run under deterministic workflow controls. The internal action guard is still shadow/observe, and we publish the failures and the current mode.
LEADERSHIP — MEET THE FOUNDER
Irby Thompson, CEO & Founder of containment.ai
Irby Thompson
LINKEDIN ↗
CEO & FOUNDER

Irby Thompson is a national security executive and serial entrepreneur with a track record of building successful cybersecurity companies. As CEO and Founder of containment.ai, he brings decades of experience in high-assurance security architectures to the challenge of AI governance.

He previously launched and grew Pikewerks (acquired by Raytheon), Star Lab (acquired by Wind River/Aptiv), and OP[4] (acquired by Kudu Dynamics/Leidos), building technologies that addressed critical security gaps in aerospace & defense, embedded systems, and critical infrastructure security.

A Vanderbilt and Georgia Tech graduate, Irby is a recognized voice on AI security challenges — particularly the fundamental risks in LLM architectures that mix instructions and data from both trusted and untrusted sources without distinction: the core problem containment.ai solves.

THE RECORD WE KEEP IN PUBLIC
10 AGENTS
run the company, governed by the product they ship →
EVERY FAILURE
written up and published, root cause to guardrail →
ZERO
certifications claimed before they're held — the record outranks the marketing →
ONE STANDARD
we didn't write, judged against in public — AARM v1.0 →
CONTACT
Talk to a human who has read your contract clauses.
Request a 30-minute Boundary Review
Bring one AI workflow, one consequential action, or one data boundary.
MAIL10001 Georgetown Pike, #384, Great Falls, VA 22066
PHONE+1-571-600-2270