LoadLine
— Know what you are carrying.
Every organisation draws its risk line somewhere. LoadLine is where you write down where, with the evidence attached, so your exposure stops being a matter of opinion.
We take the surprise out of AI
Business-Grade AI.
Grounded in your knowledge, not the open internet. Held to the mission you set. Governed by your people.
You can accept as much AI risk as you choose. What you cannot defend is accepting risk you did not know you were carrying.
Book a private briefing20 minutes. Free.
What we actually build§ 00 — Why we are here
We have both been the person who gets the call.
The one about a system nobody had written down. It had been making decisions for months, and the first anyone heard of it was when the consequences arrived.
Nobody was reckless. Everybody was busy. The thing simply grew in a place where no one was looking, because no one had agreed whose job it was to look.
That is the moment ARQ Lab exists for: the gap between what an organization has taken on and what its leaders know they have taken on.
§ 01 — What we do
What ARQ Lab does
We make the AI you already use safer to rely on, and we build AI workers you can hold to account. We don't build your AI — that is Anthropic, OpenAI and the rest. We make it better, safer and more reliable.
Bivalent
— The referee, not the commentator.
Enforcement lives outside the actor. An agent's decision is cleared at the moment it is taken, and leaves a record someone outside the room can reconstruct.
Digital Employees
— AI workers you can hold to account.
A defined job, access limited to that job, a required reporting format, and a named human supervisor.
§ 02 — How we work
Four things we hold ourselves to.
Before impact.
We work where failure is still preventable, not in the forensics.
Independent.
We don't grade our own work.
Never stale.
A live record, not a report that was true only on the day it was signed.
Pro-speed.
Oversight built so your board can say yes faster. We never sell slowing down.
§ 03 — Why ARQ Lab
Why ARQ Lab
Two operators who have carried this.
Dave Tyson and Sandra Estok built and ran enterprise security programmes before they built instruments for AI risk.
We are not reviewing our own work.
We don't build your AI, so nothing we assess is something we sold you.
We work before impact.
Not the forensics after a loss, and not a report that was true only on the day it was signed.
§ 04 — What we build with
We don't just advise. We build the AI workers too.
Our digital employees run under the same discipline we sell. That is the point — a method we would not hold ourselves to is not a method we would sell you.
Grounded in your knowledge.
A curated body of your organisation's own vetted material, not the open internet.
Held to the mission.
A defined job and access limited to that job, so the work stays on the task you set.
Governed by your people.
A named human supervisor, and a required format for reporting back.
Checked by something that didn't write it.
The producer never signs off on its own work. A structural rule, not good intentions.
§ 05 — The problem
AI failures don't arrive as events. They form.
Organizations are deploying AI — especially autonomous, agentic AI — faster than they can govern it. When it goes wrong, the loss is usually misfiled under another name. Churn. Write-offs. Rework. A bad quarter.
In EY's 2025 survey of 975 C-suite leaders, 64% of the surveyed companies reported losing more than $1 million to AI-related risks. Among those companies that experienced losses, the average reported loss was $4.4 million. Most never saw it labeled as AI.
EY Responsible AI survey, 975 C-suite leaders, 2025.
The gap between the speed of failure and the speed of oversight is the defining risk problem of this decade. AI risk is a business decision before, during, and beyond compliance. We exist to help you close that gap.
§ 06 — Three instruments
Instruments for declared exposure.
Managing your Artificial Risk Quotient takes two things: something that sees the exposure, and something that holds the limit. We build instruments for both, alongside a workforce that runs inside them.
See it — LoadLine™
LoadLine
The standing instrument for your known AI estate. LoadLine draws the mark across the AI systems and answers your organization declares, so what you are carrying stops being a matter of opinion. It provides a legible way to examine declared exposure against stated tolerance, replacing assumptions with evidence. Independent by structure: we don't build your AI, so we're not grading our own work.
Explore LoadLine →Hold the line — Bivalent
Bivalent
Governance at the decision moment. A commentator notices the foul; a referee stops play. Bivalent is built to be the referee: autonomous agents get a clearance at the point of action, and every decision leaves a receipt you can produce on demand. The actor doesn't hold the switch. Pieces of it already run inside our own systems, because we govern our AI the way we ask you to govern yours.
Explore Bivalent →Staff it — Digital Employees
Digital Employees
Purpose-built AI workers engineered for real business functions. Grounded in a curated body of your organization's own vetted knowledge, with work cross-checked by reviewers that didn't produce it, run under the same governance we sell. We build them so failure is visible, checked, and honest.
Explore Digital Employees →These are the first three. The discipline is what stays.
Sandra Estok
Cofounder
Dave Tyson
Cofounder
ARQ Lab was founded by two people who built enterprise security teams together, spent years apart learning the problem from different sides, and came back to build this. The people behind the instruments →
§ 08 — A private briefing
Decide before impact, not after.
What we're seeing in AI failures, which path fits your situation, and a straight answer if neither does.
How it's bought
A Baseline is fixed fee, fixed scope, about three weeks — and 100% credited toward a subscription if you continue.
On whose terms
Success criteria you write. You keep the scorecard, the findings briefing, and the fix list either way.
What it touches
Nothing to install and no system access required — it runs on what your leaders declare, not on your infrastructure.
We're early, and we say so.
No customer logos on this page, because we don't rent credibility. What we have is a working instrument you can see in a briefing, a transparent method, and a commitment to independent evidence. When we have customer results, you'll see them here with receipts. Until then: judge the method.
— Dave Tyson & Sandra Estok, cofounders