All articles

Incident Management in 2026: Severity Levels, Blameless Postmortems, and the EU Reporting Deadlines

How to set severity levels, run incidents with clear roles, write blameless postmortems, and build the DORA, NIS2 and CRA reporting deadlines into the process.

The first hour of a major incident tests things no dashboard shows: who is in charge, who talks to whom, and who decides whether a regulator needs to hear about it. EU financial entities reported 3,383 major ICT-related incidents under DORA in 2025 (European Supervisory Authorities, June 2026). Reporting deadlines also reach well beyond finance now. The Dutch Cyberbeveiligingswet took effect on August 15, 2026, and Cyber Resilience Act reporting started on September 11, 2026.

Incident management is now an engineering process with legal deadlines attached. These are the questions engineering leaders most often ask about it.

Key takeaways

  • Most reported major incidents in EU finance are system failures rather than attacks, so the process has to serve ordinary engineering outages first.
  • Severity describes customer impact and sets the response, while reportability is a separate legal question that needs its own named owner.
  • An incident commander coordinates instead of debugging, and the team mitigates first and looks for the root cause afterward.
  • DORA, NIS2 and the CRA each run their own reporting deadlines, and the shortest is four hours after an incident is classified as major under DORA.
  • MTTR is a weak basis for decisions, so judge the program by the quality of its reviews and the follow-up actions it completes.
A network operations center with rows of engineers at workstations, facing a large wall of screens that show live dashboards, topology maps and status indicators.
Photo: Masdestructive, CC BY-SA 3.0, via Wikimedia Commons

What actually causes major incidents?

Mostly ordinary failures. The joint supervisory report on DORA incident data shows what EU financial entities filed for 2025 (ESAs, 2026):

51%

of major incidents were reported as system failures

10%

were cybersecurity-related

1,056

had a cross-border impact, about a third of the total

The supervisors caution that reporting practices still diverge, so treat these as a first baseline. They also found that almost one third of major incidents came from failures attributable to third parties, so your process has to work when the fix is not yours to make.

Detection is the other weak spot. In a survey of 1,039 practitioners by NeuBird, which sells AI incident tooling, 78% reported at least one incident where no alert fired (IT Brief, April 2026). That is why observability work and incident work belong together.

What are incident severity levels, and how should you set them?

Severity levels turn a stressful judgment call into a lookup. They should describe customer impact, not technical cause, and the first responder should be able to pick one in minutes. A reasonable starting point:

LevelCustomer impactResponseAfterward
SEV1A critical service is down or data is at risk.Page immediately, full incident roles, reportability check at once.Postmortem always.
SEV2A critical service is degraded, or a clear group of customers is affected.Page immediately, incident commander assigned, reportability check.Postmortem by default.
SEV3Minor or partial impact with a workaround.Handled by the owning team during working hours.Short review if it repeats.

Keep downgrading easy, so people do not hesitate to declare. Google's SRE guidance lists "Declare incidents early and often" among its basic principles (Google SRE Workbook). And never use severity as a proxy for legal reportability: a SEV3 that touches personal data can still start a clock.

What does an incident commander do?

The commander runs the response and stays out of the debugger. Google's model splits the work into three roles:

Incident commander

Leads the response, holds the overall picture and makes the calls. The other two roles report here.

Communications lead

The public face of the response: status updates, stakeholders and customers.

Operations lead

Applies operational tools to mitigate or resolve the incident.

The commander's most useful instruction is usually "stop the bleeding." Google states the order directly: it "always aims to first stop the impact of an incident, and then find the root cause." Roll back, shift traffic or disable the feature first.

How do you reduce MTTR, and should you try?

Be careful with this one. Google's Štěpán Davidovič used Monte Carlo simulation to argue that MTTR and similar statistics are "poorly suited for decision making or trend analysis" (Incident Metrics in SRE). DORA, the research program, replaced MTTR in 2023 with failed deployment recovery time, because the older definitions did not separate failures caused by a software change from those caused by external factors (DORA, January 2026).

Recovery still improves when you work on the phases. Look at individual incidents, ask where the time went (noticing, declaring, finding the right person, choosing a mitigation, applying it), fix the slowest phase you keep seeing, and rehearse with game days.

What are the EU incident reporting deadlines?

Each regime has its own trigger and deadlines. This summary draws on Lewis Silkin's October 2026 overview, the European Commission's CRA guidance and the ESAs report. Confirm your own scope with counsel.

Regime and triggerFirst reportFollow-upFinal report
DORA: major ICT-related incident at a financial entityWithin 4 hours of classifying it as major, and no later than 24 hours after becoming aware of it.Within 72 hours of the initial notification.One month after the last intermediate report.
NIS2: significant incident at an in-scope entityEarly warning within 24 hours of awareness.Incident notification within 72 hours.One month after the incident notification.
CRA: actively exploited vulnerability or severe incident in a product with digital elementsEarly warning within 24 hours of awareness.Full notification within 72 hours.Vulnerability: no later than 14 days after a corrective measure is available. Severe incident: within one month of the 72-hour notification.

GDPR runs alongside these: a personal data breach generally has to be notified within 72 hours of becoming aware of it.

The final DORA report can be a long way off

The final report follows the completed root cause analysis. ESA staff instructions say that until then they expect at least one intermediate report per month (ESAs, September 2026), so someone has to keep owning the incident after the service recovers.

How do you build reporting into incident response?

Do not leave it for the legal team to discover the next morning: the 24-hour limits may have run out by then.

  1. Map services to regimes in advance

    Record in your service catalog which regimes apply to each service, so nobody works it out during an outage.

  2. Timestamp awareness at declaration

    The first entry in the incident log captures when your organization knew, because the 24-hour limits run from that moment.

  3. Name a reachable decision owner

    The commander pages a compliance or legal contact at SEV1 and SEV2. That person decides on classification, and the engineers stay on mitigation.

  4. Collect the classification data live

    In 2025 the vast majority of incidents became major through two criteria: duration and service downtime, and clients, counterparts and transactions affected. Keep those numbers current during the incident.

  5. Rehearse with templates

    Include draft reports in your game days, so gaps in access and approvals surface before a real deadline.

How do you write a blameless postmortem?

Agree the triggers in advance. Google's list includes user-visible downtime beyond a threshold, any data loss, on-call intervention such as a rollback, and a monitoring failure. A blameless postmortem, in its words, must focus "on identifying the contributing causes of the incident without indicting any individual or team" (Google SRE Book).

Do

  • Build the timeline from logs and chat history.
  • Ask why the action made sense at the time.
  • Give every action an owner and a due date.

Avoid

  • Stopping at "human error" as the cause.
  • Actions such as "be more careful."
  • Filing the document without a review.

"An unreviewed postmortem might as well never have existed," as the SRE Book puts it. The regulatory report has a different job. A DORA final report asks you to select high-level root cause categories, including human error, while your internal review should ask what made the error easy to make.

Where to start

Take your last three serious incidents and replay them on paper. Could you say when you became aware, who was in command and which regime applied, all within the first hour? Publish a one-page severity table, name the decision owner, and run one game day that ends with a practice notification.

Sources

  1. 2025 Report on major ICT-related incidents (JC 2026 16) - European Supervisory Authorities, June 3, 2026
  2. DORA Incident Reporting - Operational Instructions - ESAs staff, September 16, 2026
  3. Cyber Resilience Act - Reporting obligations - European Commission, updated September 11, 2026
  4. Mission: reportable - A guide to the (many) cyber reporting deadlines - Lewis Silkin, October 30, 2026
  5. NIS2-update: Cyberbeveiligingswet treedt op 15 augustus 2026 in werking - Forvis Mazars, August 3, 2026
  6. NeuBird AI study finds gap on incident management AI - IT Brief, April 7, 2026
  7. Incident Response - The Site Reliability Workbook, Google
  8. Postmortem Culture: Learning from Failure - Site Reliability Engineering, Google
  9. Incident Metrics in SRE - Štěpán Davidovič, Google
  10. A history of DORA's software delivery metrics - DORA, January 2, 2026

Where to start

Start with your last release

The free Release Check shows where your last release spent its time and what that costs a year, in under 5 minutes and without a call.

Keep reading