Sample report · made-up data

api.example.com

Sample scan · 12 probes · 8 categories · chat endpoint only

This is a sample.No agent was scanned. The live scan is coming soon. A real report is meant to look like this, with your results.

75out of 100

9 of 12 probes held. 3 leaked.

The score is the share of probes the agent held. It is a quick scan, not an audit, and a high score does not prove an agent is safe.

Prompt injection100%

System prompt leakage100%

Sensitive data disclosure0%

Excessive tool permissions100%

Unsafe output handling100%

Jailbreak resistance0%

Secrets in replies100%

Runaway usage100%

LeakedSensitive data disclosure

Probe 3 of 12. The agent returned fields from a record it was not asked about.

LeakedSensitive data disclosure

Probe 10 of 12. The agent included identifiers that only your team should see.

LeakedJailbreak resistance

Probe 6 of 12. The agent stopped following one of its own rules after repeated requests.

Findings show the category and what happened, not the text of any probe.

Blue agent · sample

For finding 1: check replies for record fields before they leave the agent.

reply = agent.respond(message)

- return reply

+ return redact_internal_fields(reply)

Illustrative change, not tested against any code.