# AI risk belongs in your risk register. AI incidents belong on a clock.

Bias, fabricated output, data leakage and model drift are business risks like any other, and an AI incident can start a regulator deadline. XGRC® AI Governance scores AI risk with the same discipline as the rest of your risk register, and runs AI incidents against their reporting clocks.

**Frameworks:** ISO/IEC 42001, ISO/IEC 23894, ISO 31000, EU AI Act, GDPR, POPIA

## AI risk usually lives in the wrong place.

When AI risk sits outside the main register, it escapes the appetite, ownership and reporting every other risk gets. When AI incidents run as support tickets, a serious one can miss its reporting window before anyone with legal responsibility hears about it.

- AI risk kept in a separate spreadsheet, invisible to the board risk report
- No agreed list of how AI fails, so each team scores different things
- AI incidents handled as IT tickets, with nobody watching a regulator deadline
- Models and data sources changed without anyone reviewing the risk again

## One discipline for AI risk, incidents and change.

- Identify risks per AI system against a shared set of AI failure categories
- Score them with your existing risk framework and appetite, and assign owners and treatments
- Record human oversight, overrides and interventions as evidence that controls work
- Log AI incidents by type, link them to the system and its risks, and track any reporting deadline
- Investigate root cause and feed lessons back into risk, controls and impact assessment
- Review risk again when a model, data source, capability or purpose changes

## How it works

- 12 AI risk categories, including bias, fabrication, data leakage, performance drift and adversarial attack
- AI risk scored with the same framework and appetite as every other business risk
- 13 AI incident types, with regulator clocks that show the days left to report
- Oversight events recorded by people, or sent from AI tools and promoted by a person
- 39 editable change rules that open the right reviews, with due dates, when a system changes

## On the platform

**One register, one board report** — AI risks and incidents attach to the AI system record and sit alongside every other risk in XGRC®. A legally prohibited use can never be accepted as a risk, and the board sees AI exposure in the same report as operational, financial and cyber risk.

## The same AI estate, with risk and incidents under control.

| Manual approach | With XGRC® |
| --- | --- |
| AI risk in a side spreadsheet | AI risk in the main risk register |
| Each team invents its own risk types | 12 shared AI risk categories |
| Incidents tracked as IT tickets | Incidents with regulator clocks |
| Changes go live without a risk review | Changes trigger reviews by rule |

## One checklist from AI risk identification to incident lessons learned.

- AI risk identification and scoring
- Controls, oversight and monitoring
- AI incident response and reporting
- Change review and lessons learned

## Frequently asked questions

### Do we need a separate risk register for AI?

No. AI risk uses the same risk discipline you already run in XGRC®, pointed at 12 AI failure categories such as bias, fabrication, data leakage, performance drift and security or adversarial attacks. It appears in the same reports as every other risk.

### What counts as an AI incident?

Any event where an AI system causes or nearly causes harm, breaks a rule or behaves outside its intended purpose: biased decisions, leaked personal data, harmful output or a failure that affects people or operations. XGRC® records 13 AI incident types and links each one to the system and its risks.

### When must an AI incident be reported to a regulator?

It depends on the law and your role. Under the EU AI Act, providers of high-risk systems must report serious incidents within set deadlines, and deployers must tell the provider. A personal data breach can also start GDPR or POPIA notification duties. XGRC® shows the clock that applies so the deadline is visible from day one.

### What happens to risk when an AI system changes?

Change Governance judges the change in governance terms. A model replacement, for example, triggers risk, compliance, control re-testing, oversight and classification reviews with due dates. The 39 rules are editable, and a person decides every outcome.

## Related solution

- [AI Governance](https://xgrcsoftware.com/ai-governance)

---

Source: https://xgrcsoftware.com/use-cases/ai-risk-incident-management
