Skip to content

Expert

Securing an AI application, beyond injection

The model's output treated as an instruction, third-party tools booby-trapped, a document index that leaks between users, a bill that explodes: the vulnerabilities that remain once prompt injection has been dealt with, and how to close them in the code rather than in the prompt.

5 lessons~56 mina quiz with every lesson

Start the course in the Academy

The lessons of the course

  1. Mapping the threat landscape: the OWASP list11 min
  2. The model's output is an untrusted input11 min
  3. Les pièces que vous n'avez pas écrites12 min
  4. Le RAG qui en dit trop11 min
  5. Bornes, abus et tests d'attaque11 min

Every lesson can be read here in full; the quiz, your progress and the certificate are in the Academy.

Other expert courses

The whole Academy →The AI glossary