diff --git a/slides/exercises/netpol-brief.md b/slides/exercises/netpol-brief.md
new file mode 100644
index 00000000..89f62120
--- /dev/null
+++ b/slides/exercises/netpol-brief.md
@@ -0,0 +1,7 @@
+## Exercise — Network Policies
+
+- Implement a system with 3 levels of security
+
+  (private pods, public pods, namespace pods)
+
+- Apply it to the DockerCoins demo app
diff --git a/slides/exercises/netpol-details.md b/slides/exercises/netpol-details.md
new file mode 100644
index 00000000..20db0429
--- /dev/null
+++ b/slides/exercises/netpol-details.md
@@ -0,0 +1,63 @@
+# Exercise — Network Policies
+
+We want to to implement a generic network security mechanism.
+
+Instead of creating one policy per service, we want to
+create a fixed number of policies, and use a single label
+to indicate the security level of our pods.
+
+Then, when adding a new service to the stack, instead
+of writing a new network policy for that service, we
+only need to add the right label to the pods of that service.
+
+---
+
+## Specifications
+
+We will use the label `security` to classify our pods.
+
+- If `security=private`:
+
+  *the pod shouldn't accept any traffic*
+
+- If `security=public`:
+
+  *the pod should accept all traffic*
+
+- If `security=namespace`:
+
+  *the pod should only accept connections coming from the same namespace*
+
+If `security` isn't set, assume it's `private`.
+
+---
+
+## Test setup
+
+- Deploy a copy of the DockerCoins app in a new namespace
+
+- Modify the pod templates so that:
+
+  - `webui` has `security=public`
+
+  - `worker` has `security=private`
+
+  - `hasher`, `redis`, `rng` have `security=namespace`
+
+---
+
+## Implement and test policies
+
+- Write the network policies
+
+  (feel free to draw inspiration from the ones we've seen so far)
+
+- Check that:
+
+  - you can connect to the `webui` from outside the cluster
+
+  - the application works correctly (shows 3-4 hashes/second)
+
+  - you cannot connect to the `hasher`, `redis`, `rng` services
+
+  - you cannot connect or even ping the `worker` pods
diff --git a/slides/exercises/rbac-brief.md b/slides/exercises/rbac-brief.md
new file mode 100644
index 00000000..fe319a46
--- /dev/null
+++ b/slides/exercises/rbac-brief.md
@@ -0,0 +1,9 @@
+## Exercise — RBAC
+
+- Create two namespaces for users `alice` and `bob`
+
+- Give each user full access to their own namespace
+
+- Give each user read-only access to the other's namespace
+
+- Let `alice` view the nodes of the cluster as well
diff --git a/slides/exercises/rbac-details.md b/slides/exercises/rbac-details.md
new file mode 100644
index 00000000..5809d1e3
--- /dev/null
+++ b/slides/exercises/rbac-details.md
@@ -0,0 +1,97 @@
+# Exercise — RBAC
+
+We want to:
+
+- Create two namespaces for users `alice` and `bob`
+
+- Give each user full access to their own namespace
+
+- Give each user read-only access to the other's namespace
+
+- Let `alice` view the nodes of the cluster as well
+
+---
+
+## Initial setup
+
+- Create two namespaces named `alice` and `bob`
+
+- Check that if we impersonate Alice, we can't access her namespace yet:
+  ```bash
+  kubectl --as alice get pods --namespace alice
+  ```
+
+---
+
+## Access for Alice
+
+- Grant Alice full access to her own namespace
+
+  (you can use a pre-existing Cluster Role)
+
+- Check that Alice can create stuff in her namespace:
+  ```bash
+  kubectl --as alice create deployment hello --image nginx --namespace alice
+  ```
+
+- But that she can't create stuff in Bob's namespace:
+  ```bash
+  kubectl --as alice create deployment hello --image nginx --namespace bob
+  ```
+
+---
+
+## Access for Bob
+
+- Similarly, grant Bob full access to his own namespace
+
+- Check that Bob can create stuff in his namespace:
+  ```bash
+  kubectl --as bob create deployment hello --image nginx --namespace bob
+  ```
+
+- But that he can't create stuff in Alice's namespace:
+  ```bash
+  kubectl --as bob create deployment hello --image nginx --namespace alice
+  ```
+
+---
+
+## Read-only access
+
+- Now, give Alice read-only access to Bob's namespace
+
+- Check that Alice can view Bob's stuff:
+  ```bash
+  kubectl --as alice get pods --namespace bob
+  ```
+
+- But that she can't touch this:
+  ```bash
+  kubectl --as alice delete pods --namespace bob --all
+  ```
+
+- Likewise, give Bob read-only access to Alice's namespace
+
+---
+
+## Nodes
+
+- Give Alice read-only access to the cluster nodes
+
+  (this will require creating a custom Cluster Role)
+
+- Check that Alice can view the nodes:
+  ```bash
+  kubectl --as alice get nodes
+  ```
+
+- But that Bob cannot:
+  ```bash
+  kubectl --as bob get nodes
+  ```
+
+- And that Alice can't update nodes:
+  ```bash
+  kubectl --as alice label nodes --all hello=world
+  ```