Interface Evaluation

The security action determined by AI Guard:

• 'ALLOW': The conversation is safe to proceed
• 'DENY': The current conversation exchange should be blocked
• 'ABORT': The full workflow should be terminated immediately

The evaluated conversation, redacted when required by the AI Guard service. This may contain sensitive data when redaction is disabled or no replacement was applied.

Human-readable explanation for why this action was chosen.

The replacements the AI Guard service determined for the evaluated conversation, reported whether or not the tracer applied them. Empty when the service determined no replacement.

Sensitive Data Scanner findings from the evaluation.

Dictionary of tag probabilities (e.g. { indirect-prompt-injection: 0.2, jailbreak-attempt: 0.8 })

List of tags associated with the evaluation (e.g. indirect-prompt-injection)