The usual phases are prepare, detect and analyze, contain, eradicate, recover and review. Preparing means having a plan, a contact list, access to the right systems, logs and backups. Containing means stopping the spread, for example by isolating machines, disabling accounts and revoking tokens. Eradicating means removing the attacker's access, including any way back in. Recovering means restoring systems carefully. The review records what happened and what should change.
The hardest part is scope: knowing what else the attacker touched. A common miss is to close the way in but leave persistent access behind, such as new accounts, app secrets or mailbox forwarding rules.
Vendor incidents put the response in your hands. When Railway disclosed unauthorized access to systems holding environment variables in October 2026, the work for customers was to decide which secrets were in its custody, rotate them at the source and check logs for any use. For CVE-2026-21589, the Atlassian flaw, the response meant patching, restricting access and looking in access logs for traversal patterns.
Write a one-page plan with names and phone numbers, keep a copy where you can reach it when systems are down, practice with a tabletop exercise, and keep a running timeline during an incident.