Policy pack · Security
Data exfiltration guard policy pack
Path-dependent security manifest that prevents data exfiltration from AI agents. Once an agent reads from a sensitive data source (classified as PII), any subsequent external communication is permanently blocked for the remainder of the task.
What's in it
4 policies, 4 of them blocking.
Critical policies block the step before it runs. High, medium and low ones let it run and open an incident.
| Policy | Checked | Severity | Rule |
|---|---|---|---|
| Block external send after sensitive data read | Before every step | critical | tainted_path_block |
| External sends require an exfiltration review gate | Before every step | critical | step_requires_gate |
| Restrict outbound to approved domains | Before every step | critical | domain_allowlist |
| Forbid credential read followed by external send | Before every step | critical | tainted_path_block |
Version 1.0.4. Source: manifests/security/data-exfiltration-guard.yaml.
Assign it
One command, no redeploy.
Assign it from the dashboard or the CLI. Running agents pick up an assignment within the policy refresh window, five minutes by default.
- Policies
- 4, of which 4 are checked before every step.
- Version
- 1.0.4
$ kyvvu list-manifests$ kyvvu assign-manifest --agent-id <agent> --repo-id <repo> \ --manifest manifests/security/data-exfiltration-guard.yamlAbout this pack
This is the clearest demonstration of why execution-history-aware governance matters: the same outbound step.message POST is allowed in a clean task but blocked after a sensitive read. A stateless per-call policy cannot enforce this.
Defense-in-depth layers: 1. Tainted path block (history-aware, permanent) 2. Gate before external sends (structural) 3. Domain allowlist (network boundary) 4. Credential-to-message sequence prohibition (path-aware)
Runtime policies in this pack
- Block external send after sensitive data read (critical)
- External sends require an exfiltration review gate (critical)
- Restrict outbound to approved domains (critical)
- Forbid credential read followed by external send (critical)
From Kyvvu's open policy library at github.com/Kyvvu/manifests, Apache 2.0.
Next
Run the whole pack in your browser.
In your browser
Try it in the playground
The policy is loaded. Add the steps your agent takes and watch the engine decide, step by step.
On your machine
Five minutes, no account
kyvvu try runs a demo agent against the bundled baseline. Then assign this pack from the dashboard or the CLI.
For security teams
See it on your agents
A 30-minute demo on the agents you actually run, with the policies that fit them.