Agent as a Kubernetes DaemonSetTA0008

Detect lateral movement

When an intruder is already inside, the traffic that matters is one workload reaching another. Give the flows names with the operator's identity producer, compile your NetworkPolicy into declared conversations, and review what was observed but never declared. Then declare it, or contain it.

Overview

Nothing crossed the perimeter, so the perimeter told you nothing. On Kubernetes, east-west traffic looks like an ephemeral IP talking to another ephemeral IP, and by the time you investigate, the pod is gone and the address reused.

The service graph resolves addresses to identity: which workload, which namespace, which app. A flow then reads as checkout talking to the payments database, not as two IPs.

When you see this

  1. A workload connects to a service it has never needed before.
  2. A conversation shows up that your NetworkPolicy does not describe.

Steps

  1. 01

    Run the agent on every node

    Deploy the agent as a DaemonSet so every node observes its own traffic. The synapse-stack chart installs the proxy by default and leaves the agent off, so turn it on in your values.

    synapse-values.yaml
    agent:
      enabled: true
    Helm
    helm upgrade --install synapse-stack gen0sec/synapse-stack \
      -n synapse-os --create-namespace \
      -f synapse-values.yaml
  2. 02

    Turn on the producers

    The identity producer maps pod IP to workload, namespace and app; the edge producer compiles NetworkPolicy into the conversations you intended. Both need an API key with identity:write.

    Chart values
    operator:
      identityProducer:
        enabled: true
      edgeProducer:
        enabled: true
      apiKeySecret:
        name: synapse-operator-api-key
    DecisionIs the chart older than synapse-stack 0.6.0?

    Yes Its ClusterRole lacks pods, namespaces and networkpolicies, so both producers fail. Upgrade, or apply the workaround in the operator docs.

    No The producers have the access they need.

  3. 03

    Review undeclared conversations

    A conversation nobody declared is the signal. It is not proof of compromise, but it is a short, reviewable list, which is more than a flow log gives you.

    DecisionIs the conversation intended?

    Yes Declare it in NetworkPolicy, so it stops appearing.

    No Treat the source as suspect: follow the contain a compromised host playbook.

  4. 04

    Enforce it, if you want to

    Detection does not stop anything. A smart firewall rule can name workloads rather than addresses: start with log, then switch to drop.

    Smart firewall rule, log first
    identity.k8s.src_workload == "checkout" and identity.k8s.dst_workload == "payments-db"

Common mistakes

  1. Running more than one cluster without --cluster-id: they share one path and the last writer wins.
  2. Expecting detection without NetworkPolicy: with no declared edges there is nothing to compare against.
  3. Relying on workload names when labels are not meaningful.

Escalate when

  • An undeclared conversation into a data store or a control plane.

Where this stops working

  • This is detection, not prevention. Enforcement still happens in the kernel, through the firewall.
  • The baseline is periodic: a pod that lives and dies inside a gap may not resolve.
Run it here

Send the request. Watch where it stops.

Edit the request or any rule and the pipeline runs again. The rules are real wirefilter expressions, evaluated the way the agent evaluates them; the traffic and model scores are samples. It all runs in your browser.

Try this Switch the rule's action from log to drop to enforce it.

The request

east-west · agent mode

Every value is a rule field. Change one and the pipeline re-runs.

Connection
Fingerprints, from Dendrite
Workload identity, from the operator
Model scores, sample output from Cortex
Flow behaviour, per source
Payload, what Thalamus inspects
Result Reached its destination
ResultReached its destinationAllowed, with findings · Undeclared conversation · Logged by Amygdala
  1. 01Capture and fingerprintDendrite · On the wireObserved

    Computed JA4T for this connection. Capture only: Dendrite never blocks.

  2. 02Access rulesHillock · XDP hook, before routingNot used

    Access rules answer whether a source may reach this host, so they apply to inbound traffic only.

  3. 03Intrusion detectionThalamus · On the hostNot used

    No signatures loaded for this scenario.

  4. 04Behavioural modelsCortex · On the host CPUObserved

    Sample scores, 0 to 100: C2 2, scan 3, malicious flow 41. Nothing leaves the machine to be scored.

  5. 05Service graphHippocampus · On the hostFlagged

    checkout → payments-db was observed but never declared in NetworkPolicy. A signal, not proof of compromise.

  6. 06Smart firewall rulesAmygdala · Decided on the host, enforced in nftablesFlagged

    Matched a log rule: recorded what it would have done, enforced nothing.

    Matched
  7. 07Firewall rulesHillock · TC hook, in the kernelNot used

    No firewall rules in this policy.

  8. 08WAF rulesProxy · In the proxy, after decryptionNot used

    Agent mode does no HTTP proxying, so there is no request for the WAF to read.

  9. 09Content scanningProxy · ClamAV, before forwardingNot used

    No WAF rule selected this request for scanning, so nothing was scanned.

Field names and operators follow the docs: WAF and Smart Firewall. A rule that names an unknown field does not load; a field the request does not carry never matches.

What lands in the event log
{
  "ts": "2026-11-17T09:41:07.212Z",
  "layer": "smart_firewall",
  "action": "log",
  "source": "microseg",
  "signals": [
    "microseg"
  ],
  "rule_id": "s9",
  "src_ip": "10.42.1.17",
  "src_port": 54321,
  "dst_ip": "10.42.3.9",
  "dst_port": 5432,
  "fingerprints": {
    "ja4t": "64240_2-4-8-1-3_1460_10"
  }
}

Shaped like the documented event schema; the timestamp, ports, rule ids and ban duration are illustrative.

Now on your own traffic.

A demo runs the same pipeline against real packets, with the engineers who built it.