Home/For responders/Cloud detection and response
Cloud detection and response
What a cloud responder needs to know and have in place, and where this site covers each part.
Cloud incidents are decided by what was set up before they started: logs nobody can delete, evidence that survives autoscaling, keys that can be switched off, guardrails that refuse the dangerous call. Responders rarely build these themselves. They advise the teams who do, and they need to know what good looks like, what to ask for, and what each piece records when something goes wrong. This track is organised around that work, in 10 areas.
Leading a cloud incident
Detecting, scoping, containing, preserving, eradicating, and recovering across cloud accounts, Kubernetes, and pipelines, with one person clearly in command and leadership kept informed.
On this site
- Leading a cloud incident: phases, containment per cloud, and command
- Running a cloud incident as an organisation: roles from the Incident Command System, operational periods, disclosure clocks, NIST SP 800-61r3, and a maturity ladder
- Evidence that survives termination: preserve and isolate with one tag, on four clouds
- Audit logs nobody can delete, on four clouds
Evidence that survives
Disk and memory captured automatically at the moment of containment, so evidence from short-lived and autoscaled machines outlives the machine.
On this site
- Automated snapshots, AVML memory capture, and a locked evidence bucket
- evidence-verify, a Go tool that checks downloaded evidence against the hashes recorded at capture
- Order of volatility, and why the snapshot comes first
- Keeping a compromised pod alive for investigation
- Evidence from a compromised container on GKE, EKS, AKS, and OKE, captured by one script
- Container forensics: where the evidence lives and the order to capture it
- Incident response on Google Cloud: audit log defaults, identities, VM forensics
- Incident response on AWS: CloudTrail defaults, keys and role sessions, EC2 forensics
- Incident response on Azure: Activity Log and Entra logs, managed identities, VM forensics
- Incident response on Oracle Cloud: the Audit service, principals, instance forensics
- An analysis machine that cannot reach production
Kubernetes and container runtime
Investigating node and cluster compromise, container escape, admission bypass, and workload identity abuse, at the pod, node, and API server.
On this site
- Kubernetes for defenders: attack paths and the telemetry each one leaves
- Containers on GKE and EKS, including Autopilot, Fargate, and Auto Mode
- Containers on AKS and OKE, including virtual nodes and AKS Automatic
- Clearing an attacker out of a cluster: certificates, tokens, webhooks, and CA rotation
- Falco runtime detection, with rules in Git
- Argo CD, where drift from Git becomes visible
- Stop risky pods before they start
- A record of every action against the cluster API
- Encrypt and authorise every call between services
Telemetry you can hunt in
The cloud's own audit logs, network flow logs, and Kubernetes audit logs, collected, kept, and turned into detections.
On this site
- Control-plane audit logs kept where attackers cannot reach them
- Host logs sent off the machine as they are written
- What to look for first in each cloud's audit log
- Keep network records long enough to be useful
- Get cloud logs into your SIEM
- A record of every action against the cluster API
Identity and guardrails to advise on
Roles with only the access they need, the ways one identity can become another, pipeline sign-in without stored keys, and rules set above the account. All of it reviewed and explained to the teams who own it.
On this site
- Reviewing infrastructure as a defender: trust policies, impersonation, and perimeters
- Encryption keys with managers and users kept apart
- How each cloud names identity, keys, and audit
- Block the changes you never want
- Cloud access for pipelines and pods without stored keys
- Data that cannot leave your organisation
Pipelines and supply chain
Leaked secrets, compromised builds, risky dependencies, and proof of where software came from, with pipeline findings reaching responders.
On this site
- What to check in a pipeline and a pull request
- This site's own pipeline, as a worked example
- Deploying by pull, so the build system has no cluster credentials
- Pipeline security findings in front of responders
- Cloud access for pipelines and pods without stored keys
- Stop risky pods before they start
Hosts, in and out of the cloud
Investigation across Linux, macOS, and Windows, including the hosts under cloud workloads.
On this site
- SSH locked to keys on Linux and Windows, with the setting read back to prove it
- Host logs that survive the host: rsyslog on Linux, Sysmon and Event Forwarding on Windows
- auditd rules for accounts, sudo, SSH keys, persistence, and root commands
- osquery on Linux, macOS, and Windows, logging what changed
- Falco on servers outside Kubernetes
- Host forensics on Linux, macOS, and Windows: what to capture first, persistence, and logs
- Linux memory capture with AVML
- An analysis machine that cannot reach production
Insider cases
Recognising the steps someone with legitimate access takes on the way out with data, and working the case with HR and Legal so the findings stand up.
On this site
- Working an insider case: triage, authority, evidence, and hand-over
- Alerts on sharing outside, bulk reads, decrypt rights, and logging turned down
- Windows audit policy that records copies to removable drives
AI and LLM systems
Keeping a record of who called which model, and spotting stolen credentials used to run models on someone else's bill.
On this site
- Working a case of stolen credentials running models: attribution, containment, the bill, and the data exposure
- Model call records, and alerts on logging, guardrails, keys, and new deployments, on four clouds
- Which AI service logs exist, which are off by default, and how long each is kept
Teaching the next responder
Raising a team's cloud and container fluency, and turning each incident's findings into changes the owning teams make.
On this site
- Plain-language lessons on every tool, with diagrams and real output
- Common mistakes, by tool
- Turning a review's findings into guardrails
Companion sites
Detection content, hunting, and AI security live on their own sites.