SECAI Core · phase 11 of 15
Improper output handling
When generated responses are used directly in applications without verification
The Explain card
- Plain English
- Improper output handling is using a model's response directly in an application without checking it first: rendering it, executing it, or passing it to another system as though it were trusted.
- Example
- A web app drops an LLM's answer straight into the page as HTML. An injected prompt makes the model return a script tag, and the browser runs it, stealing the user's session.
- Why it matters
- Model output is untrusted input to everything downstream. Encoding, validation, and sandboxing between model and consumer stop prompt injection from becoming XSS, SQL injection, or command execution.
- Hook
- The model's answer is a stranger's input with good grammar.
Where it sits in the deck
Phase 11: Threat Landscape: Attack Vectors and Adversarial Techniques
With defences named, learn what they must defend against — the full catalogue of attack techniques targeting AI systems, their inputs, outputs, training pipelines, and supply chains.
