Skip to content

Security

Responsible disclosure

We welcome reports from security researchers and will work with you to fix real issues quickly.

How to report

Email security@vigiltrism.ai (or support@vigiltrism.ai) with steps to reproduce, the affected address and the impact you observed. We acknowledge reports within 48 hours. Reports in English or Hebrew are welcome.

In scope

vigiltrism.ai, the dashboard and the public API under /api/public/v1.

Please don't

  • Access, change or delete data that isn't yours.
  • Run denial-of-service, spam or social-engineering tests.
  • Publicly disclose an issue before we've had a reasonable chance to fix it.

Our commitment

We won't pursue legal action against good-faith research that follows this policy, and with your permission we'll credit you once the fix ships. We don't currently offer paid bounties.

Data handling

  • Pre-flight scans in memory: inputs sent to the scan API are evaluated in memory and allowed inputs are not stored.
  • No training on your data: we never use customer prompts, outputs or telemetry to train models, and never sell or share them.
  • Hashed keys, scrubbed traces: agent API keys are stored only as SHA-256 hashes; credentials are scrubbed from execution traces before storage, and traces are kept for your plan's retention window.
  • Encrypted in transit: all traffic is served over HTTPS with HSTS.

Deployment

Vigil is a managed cloud service. Organizations that cannot send data to an external API can discuss private (in-VPC) deployment through an Enterprise inquiry on the pricing page.

Security architecture

  • Fail-closed clients: our Python and Node examples block the action on network errors, server errors or unknown verdicts.
  • Agent keys: stored only as SHA-256 hashes and shown once; credentials are scrubbed from stored payloads.
  • Workspace isolation: every table is protected by row-level security tied to workspace membership and role (Admin, Developer, Auditor).
  • Outbound requests: agent and alert webhook URLs are checked to block private networks and cloud metadata addresses; alert payloads are HMAC-signed.
  • Browser hardening: strict Content Security Policy, HSTS, X-Frame-Options DENY and nosniff on every response.
  • Model data: customer data is never used to train models. The optional AI-assisted check uses a provider endpoint configured for zero data retention.

Current status at a glance

ItemStatus
SOC 2 Type IIIn preparation — not audited
ISO 27001Control mapping — not certified
Independent penetration testNot yet performed
Vulnerability disclosureActive — 48-hour acknowledgement
Public detection benchmarkPublished with raw data

Compliance roadmap

  • SOC 2 Type II: internal control preparation in progress; we are not yet audited or certified.
  • ISO 27001: we are mapping controls into our development lifecycle; not certified.