Back to Vulnerability Hub
HighAI-Generated Code Vulnerability
Prompt Injection
Untrusted input manipulates model behavior, triggering unauthorized actions or data disclosure.
Affected Platforms
ClaudeCursorLovableCustom AI agents
Severity
High
Remediate immediately
How to detect it
- 1Inject adversarial prompt patterns in user-controlled fields
- 2Evaluate tool-call policy bypass attempts
- 3Monitor output for policy-breaking actions
How to fix it
- 1Isolate instructions from user content
- 2Use strict tool allowlists and argument validation
- 3Add post-generation policy filters
Check if your app is affected
VibeMeds automatically detects this and 41 other vulnerability patterns.