gws-modelarmor-sanitize-prompt

Prompt injection is a malicious technique where user inputs trick AI models into ignoring safety guidelines. Sanitizing user prompts prevents prompt injection by filtering harmful content and enforcing predefined security policies before the AI model processes the query.

To sanitize user inputs for AI safety, apply predefined Model Armor templates by passing the user text and template path, such as 'projects/P/locations/L/templates/T', to cleanse the prompt of malicious content before execution.

You need to sanitize prompts to ensure content safety whenever your AI application processes user-generated text, particularly before sending queries to sensitive AI systems to prevent users from revealing confidential information or performing unintended actions.

Yes, this prompt sanitization approach works for securing AI model inputs across various applications requiring user-generated text. It enforces predefined security rules via Model Armor templates to ensure content safety regardless of the specific AI application.

The best way to secure AI model inputs against malicious user queries is to sanitize the text using Model Armor templates. This approach applies predefined security policies to cleanse the input, preventing prompt injection and ensuring content safety before processing.