What is an AI risk register?
An AI risk register is the written record of the AI risks an organization has decided to accept, each with a named owner and a date it will be looked at again. Its purpose is not to list everything that could go wrong. It is to separate a risk you took deliberately from one you never noticed, because afterwards the two are indistinguishable.
Accepting a risk is a decision, and decisions can be shown
Running an AI system with a known limitation is often the right call. The output is good enough, the failure mode is recoverable, the alternative costs more than the exposure. None of that is a governance failure.
It becomes one at the point where nobody can show the decision was made. A risk that was weighed and accepted and a risk nobody thought about produce the same outcome when it materializes, and the only thing distinguishing them is a record written beforehand. That is the whole job of the register.
The two fields that make it work
An owner, and a review date. Without an owner, an accepted risk belongs to the meeting that accepted it, which is not an entity that can be asked about it later. Without a date, acceptance is permanent, including after the thing that made it acceptable changed.
Most registers fail on the second. The entries are real, the reasoning was sound, and the dates passed a year ago, so the register now describes a set of decisions made under conditions that no longer hold. A register with stale dates is worse than a short one with live ones, because it looks like coverage.
What belongs in it and what does not
It is not a catalog of everything that could go wrong with AI. A register that tries to be one gets long, gets skimmed, and stops being read at all.
What belongs is the risks somebody actually decided to carry: this system produces output nobody checks, this vendor holds data we would rather it did not, this process depends on a tool with no contract behind it. Each of those is a live decision with a name against it, and the register is where you go to find out whether it is still the decision you would make.
Common questions
- How is a risk register different from an impact assessment?
- An impact assessment looks forward at one system before it is deployed and asks what it could do to people. A risk register looks across everything already running and records what you chose to live with. The assessment is per system and mostly one-off. The register is continuous and its value is entirely in being current.
- How many entries should it have?
- As many as there are decisions you made, which for most organizations is a short list. Length is not a measure of thoroughness here. A register of eight live decisions with owners and dates is worth more than sixty entries nobody has looked at since they were written.
- Who reviews it?
- The named owner of each entry, on the date against that entry. A single review of the whole register on one day a year tends to become a formality, because the entries were accepted under different conditions at different times and none of those conditions changed on the same date.
Find out where you stand
The assessment on this site asks 22 questions about what AI is in use, what rule governs it and who is accountable, and returns a score, a verdict and, where a control is short, the one thing to fix first. It names no framework, and it takes about ten minutes.
It is a self-assessment of what you could show today, not an audit, not a certification, not a conformity assessment and not legal advice.
Run the assessment