Incident Response¶
What happens when a security incident is found in this service: who declares it, how fast it is communicated, and to whom. Covers vulnerabilities found in the code, exposure of data the service holds, and compromise of a credential or key.
Service outages that are not security incidents follow the Disaster Recovery Plan instead. An outage becomes an incident the moment data or credentials are implicated.
Reading time: ~4 minutes
Who declares an incident¶
The DR team lead named in the Disaster Recovery Plan declares an incident and owns the response, using the same roster as recovery so the two never drift apart. Anyone may raise a suspected incident; declaring it is what starts the clock below.
Severity and communication timeframes¶
| Severity | Trigger | Notify the team | Notify EPFL DSI / ENAC-IT |
|---|---|---|---|
| Critical | Data exposure, authentication or authorization bypass, key compromise, service down | Immediately | Within 24 hours |
| High | Exploitable vulnerability with no evidence of exploitation | Same business day | Within 1 week |
| Low | Hardening gap with no exploit path | Next planning cycle | In the periodic review |
Alerting already delivers the first hop: an alert reaches the sysadmin group by email roughly 30 seconds after firing, and detection runs 1–10 minutes behind the event itself. See Recovery objectives.
Personal data¶
The service stores personal data — headcount entries ingested from Tableau hold a member name, institutional identifier and FTE. It stores no sensitive personal data in the sense of the Swiss FADP: no health, biometric, genetic, religious, political or proceedings data. That distinction limits the obligations that follow, and it is the first thing an auditor will ask, so state it rather than leaving it implied. If an incident exposes them, notify the EPFL Data Protection Officer (dpo@epfl.ch) within 72 hours of becoming aware, in addition to the timeframes above.
⚠️ Do not state that this service processes no personal data. It does. Any questionnaire answer claiming otherwise is wrong and discredits the rest of the submission.
Confidentiality¶
Incident communications are restricted while the incident is open: the DR team and EPFL DSI only, and the tracking issue stays private until a fix ships. After the fix is deployed the details may be shared or published.
This is coordinated disclosure, not secrecy — the constraint is "not before the fix", not "never".
Procedure¶
- Report the suspected incident to the development team immediately.
- Declare and classify. The DR team lead assigns a severity from the table above, which sets every deadline that follows.
- Contain. Revoke or rotate an exposed credential first — rotation blast radius is documented in Encryption and Key Management.
- Fix. Critical findings block all other work and ship as soon as a fix is verified; lower severities follow the normal release path.
- Notify EPFL DSI within the timeframe for the severity, and the
Data Protection Officer within 72 hours if personal data is
involved. Open the ticket directly in the failing service's
queue (
SI_POSTGRESQL,SI_KUBERNETES,SI_AD, …) — see the service map. - Record. Log the incident, its resolution and its lessons in a private tracking issue, and update whichever documents the incident proved wrong.
After the incident¶
Every incident ships a regression test that would have caught it, the same rule as any bug fix. If the incident revealed a documentation gap, the fixing PR updates the document — a lesson recorded only in an issue is a lesson lost.