Skip to main content

AI risk management

What could go wrong, how likely it is, how bad it would be, who owns it, and whether the evidence says it is getting better or worse.

AIC.Karoo Lending (Demo)

Risk register

Sample data

  • Unfair outcomes across groupsTreating
  • Reviewers approve without lookingOpen
  • Supplier changes the model unannouncedOpen
  • A leaver still has accessOpen

Built to answer to

  • ISO/IEC 42001 clause 6.1
  • EU AI Act Article 9
  • King V
  • NIST AI RMF

Scored the usual way

Likelihood and impact from one to five, an inherent score, a target once treated, and a current score that rises when the evidence behind a risk is failing.

See it in a walk-through

Unfair outcomes across groups

Sample data

  • Inherent16
  • Current12
  • Target6

Likelihood times impact, each from one to five

A library to start from

Ready-made scenarios for AI decisions about people, such as unfair outcomes across groups or reviewers who approve without looking, alongside the core security risks.

See it in a walk-through

Risk library

Sample data

  • Unfair outcomes across groupsAI decisions about peopleAdd
  • Reviewers approve without lookingHuman oversightAdd
  • Model drift goes unnoticedMonitoringAdd
  • No route to challenge a decisionCorrectionAdd

AIC notices first

Failing checks, undeclared systems, leavers who still have access and model retirements are raised as suggested risks, with the evidence, for you to add or dismiss with a reason.

See it in a walk-through

AIC noticed

Sample data

From your connected systems

  • Accounts can sign in without a second factorMicrosoft 365Suggested
  • A leaver still holds 3 accountsBambooHR and Microsoft 365Suggested
  • AI used by a system not on the registerOpenAISuggested

And also

  • Owned and reviewed

    Every risk has an owner and a review date set by its score. Overdue reviews and risks without an owner are counted where you will see them.

  • Tied to the system

    A risk can name the AI system it concerns, so each system on the register shows its open risks, and each risk shows its system.

What it answers to

  • ISO/IEC 42001 clause 6.1: actions to address risks and opportunities, including AI risk assessment
  • EU AI Act Article 9: a risk management system for high-risk AI
  • King V: risk governance by the governing body
  • NIST AI RMF: the Map and Manage functions

The platform is a tool. Using it neither raises nor lowers the chance of AIC certification, which is decided only against the published standard. See impartiality.

Questions

AI risk management: what does it do?

A risk register built for AI decisions about people: likelihood and impact, owners, treatment, review dates, and risks raised from what your connected systems show.

Which laws and standards does it help with?

ISO/IEC 42001 clause 6.1: actions to address risks and opportunities, including AI risk assessment. EU AI Act Article 9: a risk management system for high-risk AI. King V: risk governance by the governing body. NIST AI RMF: the Map and Manage functions.

Does using it make AIC certification more likely?

No. The platform is a tool. Using it neither raises nor lowers the chance of certification, which is decided only against the published AIC standard, by assessors who treat evidence the same way wherever it came from.

See your own AI estate on the record

A short session on your own systems, with the person who will answer for them.