Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Walkthrough — kubernetes-privesc
Namespace SA bound to
cluster-admin. A namespace-scoped ServiceAccount in theexperimentsnamespace is bound to thecluster-adminClusterRole by[email protected]at 02:14 UTC — outside the standard change-management window. The SA then immediately listssecretsinkube-system.
- Fixture:
examples/alerts/kubernetes-privesc.json - MITRE techniques: T1098, T1078
- Severity at intake:
critical - Confidence band (sandbox): 100/100,
high
Run it
# Offline simulator:
aisoc-sandbox demo --scenario kubernetes-privesc
# Real stack (after `make up`, with `pip install -e packages/aisoc-cli`):
aisoc submit examples/alerts/kubernetes-privesc.json
What the agent does, step by step
Step 0 · DetectAgent · detect
Two native detections fire: k8s-clusterrolebinding-cluster-admin
(any new binding to cluster-admin outside a change-management
window) and k8s-secrets-list-by-sa (a ServiceAccount listing
kube-system secrets). Fusion lifts to critical because of the
cluster-admin blast radius.
Step 1 · TriageAgent · triage
The graph walk identifies that [email protected] has not
performed any admin RBAC actions in the last 30 days, that the SA
debug-runner was created less than 24 h prior, and that the
namespace experiments has no production workloads — three
independent signals consistent with credential-takeover-then-privesc.
Decision: Confidence high (100/100).
Step 2 · HuntAgent · hunt
Sweeps the last 24 h for every action by [email protected]
across the cluster, and every action by the new SA debug-runner.
The hunt surfaces the secret-list call as the only follow-on action
so far — the actor hasn't yet exfiltrated anything from kube-system.
Step 3 · RespondAgent · respond
Proposes containment that breaks the privilege chain without taking the cluster down:
k8s.clusterrolebinding.delete({"name": "debug-runner-cluster-admin"})k8s.serviceaccount.disable({"namespace": "experiments", "name": "debug-runner"})user.session.revoke_all({"user": "[email protected]"})case.create(...)at severitycriticalcase.notify({"channel": "pagerduty", "service": "kubernetes-on-call"})
Decision: 5 actions proposed; awaiting analyst approval (severity critical).
What the analyst would do next
- Approve the
clusterrolebinding.deletefirst — it cuts the actor's privilege immediately. The SAdisableand the user session revoke can follow in any order. - Audit every secret in
kube-systemthat the SA could have read in its short window of cluster-admin access. Rotate anything sensitive. - Open a separate ticket on the change-management process: a
binding to
cluster-adminshould require a PR + two approvers, not a single API call. - Review the
experimentsnamespace's IAM bindings — there might be other latent SA-with-broad-perms patterns to clean up before they become incidents.