Home/Refreshers/Leading a cloud incident: the first hour: refresher
Leading a cloud incident: the first hour
The same phases as any incident, with evidence that disappears faster, more ways for an attacker to come back, and most actions taken through APIs.
Remember
- Preserve before you contain. Terminating a machine destroys the evidence.
- Most cloud intrusions are valid credentials, so the audit log is the main evidence.
- Scope what the attacker created as well as what they used: keys, users, roles, trust changes.
- One commander decides. Someone else does the hands-on work.
- Agree in advance which containment actions need nobody's approval.
The first hour
- 1. Command Name the incident commander. Open the decision log: what, who, when, on what evidence.
- 2. Preserve Tag affected instances incident=<id>. Export logs close to their retention limit.
- 3. Scope Search the audit logs for new keys, users, roles, trust policy edits, and impersonation.
- 4. Contain Disable keys, revoke sessions, isolate machines. See the containment refresher.
- 5. Report Update leadership on a fixed schedule: what we know, what we are doing, what we need, next update.
Watch out
- Revoking one key changes nothing if the attacker already made another.
- A security group change can leave open connections running. Add a network ACL deny when they must stop now.
- Google Cloud records service account impersonation only if Data Access logs are on for the IAM credentials API.
Check it worked
Before calling it contained: every credential and identity the attacker created is found and disabled, and the evidence is hashed and locked.