Security and data handling
What Leap360 sees, what stays in your process, and who can change what.
What stays in your process
- Classification of every tool call runs locally.
- The full result of a wrapped tool call is scanned locally for prompt injection and credentials.
- The code scan runs on your machine or CI runner. Only
--reportuploads its summary.
What Leap360 receives
- The action: the tool name, its arguments, the agent and the decision.
- The outcome: ok or error, and how long it took.
- A redacted preview of the result. Secrets, card numbers and emails are masked before it leaves your process.
- For traced model calls: the model, token counts, latency, and the request and response you choose to send.
Who can change what
| Change | API key | Signed-in person |
|---|---|---|
| Stop an agent or the project | Yes | Yes |
| Switch to enforce, add a policy | Yes | Yes |
| Restart an agent or the project | No | Yes |
| Turn on shadow mode | No | Yes |
| Approve or reject a held call | No | Yes |
| Change or delete a policy | No | Yes |
Stopping an agent also stops the API keys it used.
Evidence
Every decision is recorded. Each row in Evidence carries a SHA-256 hash of its stored record, so you can check later that it has not changed.
Report a vulnerability
Email security@leap360.ai.