[Feature Request] Vision-Language Model Guardrails for Sensitive Image Detection (PII, etc.)

[saleiro]

A guardrail for this would:

Detect PII in Images: Scan images for visual PII indicators.
Flag and Alert: Raise an alert if potentially sensitive content is found, providing details about the type of PII discovered and its location within the image.
Optionally Redact: Offer the ability to automatically redact or blur sensitive portions of the image before further processing or storage.

[Pouyanpi]

@saleiro Multimodal rails were added in 0.13.0 release, have a look at the release notes.