A checklist tells you what you have. The Cube tells you how much risk is left.
Most AI governance reviews end in a maturity level or a list of ticks. Neither tells a risk committee whether exposure is rising, or tells an auditor what stands behind the answer. The AI Risk Cube gives both of them a number they can trace.
- Measure
- One score for each AI system and one for the whole organization, on a 0 to 1,000 scale a board can read in seconds. Higher is better.
- Locate
- Every score breaks down into cells: which stage of the AI's life, which kind of risk, which safeguard. You see exactly where the gap is.
- Prove
- Each answer carries a proof grade, from self-declared to independently checked. Old evidence loses value until you renew it, so the score stays honest.
- Act
- Every gap comes with a control you can assign to an owner. The Cube ranks the fixes by how many points each one adds.
Risk has momentum. Your controls are the friction.
An AI system that can do serious harm, and acts fast on its own, carries momentum. Safeguards you can prove slow it down. What's left is your remaining risk, and the Cube Score shows how much of it you have under control. Move the sliders to see it work.
654, Exposed. Some of this system's risk is not well held back yet.
A simplified version of the scoring method, using fictional examples. The full method rates potential harm and speed from eight factors and Control Strength across six risk areas.
AI risk has more than one dimension.
A single risk rating flattens AI risk into one line. In practice it depends on where the system is in its life, what kind of harm is in play, and which safeguards hold it back. The Cube keeps all three dimensions, so every cell answers one precise question: at this point in the AI's life, for this kind of risk, is this safeguard in place, and can you prove it?
When in the AI's life?
Six lifecycle stages
- L1Plan and design
- L2Data
- L3Build or buy
- L4Verify and validate
- L5Deploy
- L6Operate, monitor and retire
What could go wrong?
Six risk areas
- D1Data sources and rights
- D2Data quality and governance
- D3Performance and reliability
- D4Security
- D5Fairness and impact on people
- D6Autonomy and human oversight
How is it held back?
Six kinds of safeguard
- A1Govern
- A2Inventory and classify
- A3Measure and test
- A4Control and mitigate
- A5Monitor and respond
- A6Assure and report
Six by six by six makes 216 cells. The free Scorecard folds them into 27 questions, so a free result maps straight onto the full Cube.
The fourth dimension is time.
AI doesn't hold still, so its risk doesn't either. Some risks are built in by the people who design a system. Others emerge from the system itself, as it learns, adapts and acts on its own. A cube that was accurate last quarter can be wrong today.
- Evidence ages
- Proof loses half its value after a set period: 3 months for monitoring, up to 24 months for policy. Your score drifts down until you renew it.
- Systems change
- The Cube rates how often a system's model, data, prompts or tools change, and how freely it acts. Faster and freer means more risk to hold back.
- Unreviewed change counts
- When changes go live without review, the evidence behind your controls ages faster.
Your Cube Score is a live reading, like a gauge on a machine. It moves when your AI moves.
The Cube draws on published frameworks including the NIST AI Risk Management Framework, ISO/IEC 42001, the EU AI Act and OWASP guidance for AI agents. It is not affiliated with or endorsed by those bodies.
Start free with your whole organization. Go deeper system by system.
Free Scorecard
For a first reading of your organization. Opens soon.
- 27 plain-language questions, about 10 minutes
- Your Cube Score and band, on screen right away
- The main reason behind your score
- The three actions that would raise it the most
- Full details by email: results by area and the complete action list
Cube Pro
For assurance work on each AI system. In pilot with a small group of organizations.
- A score for every AI system, rolled up to the organization
- Up to 216 cells per system, with depth matched to the risk
- Proof grades and evidence dates behind every answer
- A control roadmap ranked by points gained
- The legal duties that likely apply where you operate
- A one-page board summary and an evidence log for auditors
AI rules now come with dates.
Regulators have stopped publishing principles and started setting deadlines. And AI that acts on its own, without a person approving each step, raises the stakes. The Cube scores autonomy as a risk area of its own: what an agent can do, who approves it, and whether you can stop it.
- Texas. The Texas Responsible AI Governance Act is in force, enforced by the Texas Attorney General. Source
- Colorado. Senate Bill 26-189 on automated decisions takes effect, with notice and human-review duties. Source
- European Union. High-risk duties under the AI Act start to apply to standalone AI systems. AI inside regulated products follows on 2 August 2028. Source
Dates checked on 15 November 2026. This is general information, not legal advice.
Built for the people who answer for AI.
- Risk and compliance leadersSet risk appetite, compare systems and track the trend.
- Internal auditScope audits and test what management asserts, with an evidence log to sample.
- AI governance leadsRun intake, risk tiering and approval gates from one inventory.
- Security leadsSize the attack surface of models and agents.
- System, data and model ownersAnswer in plain language, then close the gaps assigned to you.
- Boards and executivesOversee and fund from a one-page summary: score, trend and top actions.
What the Cube is not.
- Not a certification or legal opinion. It supports readiness work for standards and laws. It does not confer conformity, and your counsel confirms what applies to you.
- Not a testing tool. It doesn't run bias or security tests. It records the tests you run, grades the evidence and dates it.
- Not finished. The scoring weights and bands are starting values. Pilot organizations are helping to calibrate them, and the method will be published.
Who is behind it.
- Master of Legal Studies (MLS), Risk Management & ComplianceTexas A&M University School of Law
- PhD candidate, Artificial IntelligenceCapitol Technology University
Luis is an operational excellence leader who also facilitates online courses for eCornell. His work joins the two sides of AI risk: how the systems behave, and what the rules require.
His doctoral research examines the risks that AI systems generate on their own as they learn, which he calls organic risks, alongside the synthetic risks that people build in.
He built the AI Risk Cube so that risk, audit and governance teams can measure AI risk, compare it and act on it, the way engineers do with any system that can cause harm.