Skip to content
Sanitized AI Help Center

How detection works

Where scanning happens

A local guard runs alongside your AI tool and inspects each prompt. When it needs deeper analysis, the prompt is sent over an encrypted (TLS) connection to your organization’s isolated detection backend.

What we detect

Detection combines pattern matching with named-entity recognition to catch:

  • Personal data (names, emails, phones, addresses, government IDs)
  • Secrets and credentials (API keys, tokens)
  • Confidential business terms your admin configures

What we log

We log metadata about detections — for example, that a prompt was flagged and what category was matched — not the content of your prompts.