The helpful agent. The hostile instruction.
Follow a prompt injection from innocent-looking content to an action you never asked for. See why “please don’t” isn’t a defense.
Prompt injectionAn AI security workshop
Let it build.
Not break in.
Your AI agent can ship code. What else can it do?
Spend an evening finding out, before someone else does.
More autonomy.
Less blind trust.
The lights are on.
Who has the keys?
Inside the workshop
Trust is not
a security boundary.
Repos, docs, dependencies. Your agent reads a lot of things you didn’t write. Let’s look at what happens when one of them talks back.
Real attack paths. Practical defenses.
No vendor pitches.
Give it a workspace. Not your whole machine.
Explore isolation, permissions, and network boundaries. Decide what your agent actually needs, and what should stay out of reach.
Agent isolationUseful access. Without the master keys.
Use piso to explore a credential gateway: the agent gets placeholders, while real secrets stay outside its workspace. Test the limits, too.
Credential securityOne event. A better threat model.
Keep building.
Bring boundaries.
A hands-on evening for developers and curious builders. Bring your questions about the agents you’re already running.
- When
- To be announced
- Where
- To be announced
- Format
- One-off workshop
Booking opens soon.
The date and location are still being confirmed. Join the interest list and we’ll send you the booking link when it’s ready.