T05 AI Incident and Breach Reporting Workflow (EU)
Verification required. This is a working template, not a regulator-approved form. Where it cites an article or a date, the citation comes from secondary sources or from memory of the legal text and is a pointer to check. This is not legal advice.
How to use: This is a blank form. Copy it and fill in the empty cells and blanks for your system. To copy it, follow the "Suggest an edit to this page" link in the footer to reach its Markdown source.
Wiki control: AI-CTRL-035. One incident can trigger several regimes at once. The timings below are [Recalled]. Confirm every deadline from the primary text and the national transposition before you use this workflow. Where a cell is blank, fill it from the source.
1. Incident types
| Code | Type | Example |
|---|---|---|
| I1 | Personal data breach | Prompt logs exposed |
| I2 | Serious incident under the AI Act (high-risk) | Death, serious harm to health, serious and irreversible disruption of critical infrastructure, serious infringement of fundamental rights, serious harm to property or the environment |
| I3 | Cybersecurity incident (NIS2 or DORA entity) | Compromise of the AI platform |
| I4 | Vulnerability in a product with digital elements (CRA) | Exploited flaw |
| I5 | Harmful or unfair output | Discriminatory decisions |
| I6 | GPAI systemic-risk incident | Provider reporting to the AI Office |
| I7 | Supplier incident | Silent model change, breach |
2. Regime matrix
Complete this matrix before an incident happens.
| Regime | Who reports | To whom | Trigger | Deadline | Source checked and date |
|---|---|---|---|---|---|
| GDPR breach | Controller; the processor notifies the controller | Supervisory authority; affected persons if the risk is high | Personal data breach with risk | 72 hours to the authority [Recalled] | |
| AI Act serious incident (Art. 73) | Provider of a high-risk system | Market surveillance authority of the Member State | Serious incident | 15 days in general; shorter for a widespread infringement or critical infrastructure; shorter for a death [Recalled] | |
| AI Act deployer duty | Deployer | Provider and authorities | Serious incident identified | Without undue delay [Recalled] | |
| GPAI systemic risk (Art. 55) | Provider | AI Office | Serious incident | ||
| NIS2 | Essential or important entity | CSIRT or competent authority | Significant incident | Early warning in 24 hours; notification in 72 hours; final report in one month [Recalled] | |
| DORA | Financial entity | Financial supervisor | Major ICT incident | Reported as an initial notice within hours; verify | |
| CRA | Manufacturer | CSIRT and ENISA | Actively exploited vulnerability or severe incident | Reporting applies from 11 Sept 2026 [Reported] | |
| Contracts | Supplier or customer | Counterparty | Per contract | Per contract |
3. Workflow
| Step | Action | Owner | Target |
|---|---|---|---|
| 1 | Detect and log | Anyone / SOC | Immediately |
| 2 | Triage: type, severity, systems, data, Member States | Incident lead | |
| 3 | Contain: pause, kill switch (AI-CTRL-025), revoke keys | Engineering | |
| 4 | Preserve logs, prompts, model version, configuration | Security | |
| 5 | Assess duties across all regimes in section 2 | DPO / Legal | |
| 6 | Notify authorities within the deadlines | Legal / DPO | Per law |
| 7 | Notify customers, suppliers, affected persons | Account owner / DPO | Per law and contract |
| 8 | Remediate and retest | Engineering | |
| 9 | Update the risk assessment, technical file and post-market plan | AI governance lead | |
| 10 | Record the incident and update T11 | GRC |
4. Severity
| Level | Definition |
|---|---|
| S1 | Harm to people or to regulated data |
| S2 | Material compliance or service failure |
| S3 | Contained |
5. Messages
Prepare these in the languages of the affected Member States: authority notice, customer notice, individual notice, internal bulletin and holding statement. Legal approves each one.
6. Record
ID, dates, type, severity, systems, persons and data, root cause, regimes triggered, notices (who and when), actions, lessons, file references.
7. Exercise
Run a tabletop exercise each year. Cover one multi-regime scenario (a personal data breach in a high-risk system) and one biased-output scenario. Keep the report as evidence.