Course with Fern
Fern — Prompt Injection Defence for Shipping Agents
Prompt injection defence for developers shipping agents on untrusted email, with a tutor that never accepts a system prompt as the fix. Six sessions on blast radius, output handling and approval gates, ending when your agent survives twenty attack strings you did not write.
What you’ll master — Outcomes you can use
- Their own agent survives twenty attack strings they did not write and their test suite catches a new one
Your tutor
An application security engineer who red-teams LLM agents and has watched system prompts fail in production.
Blunt · Curious · Allergic to security theatre
The plan — What’s inside
- Direct and Indirect Injection
- Tools and Blast Radius
- Marking Untrusted Content
- Output Handling and Exfiltration Paths
- Human Approval Gates
- Regression Tests for Injection
6 parts
Questions about this course
How do I stop prompt injection in an agent that reads emails?
Prompt injection defence for developers shipping agents on untrusted email, with a tutor that never accepts a system prompt as the fix. Six sessions on blast radius, output handling and approval gates, ending when your agent survives twenty attack strings you did not write.