Prompt Injection
High · AI Security · OWASP LLM01
Description
User text overrides system instructions in LLM apps.
Root cause
Concatenating untrusted input with prompts.
Impact
Data exfiltration, unauthorized tool use.
Detection
Red-team prompts; output monitoring.
Mitigation
Separate contexts; tool allow-lists; filters.
Secure coding
Human approval for sensitive agent actions.
Safe lab example
lab-ai-prompt
Practise related skills in Labs or deepen on Academy library.