Skip to content

Security: zgbrenner/whereas

SECURITY.md

Security Policy

Reporting a vulnerability

Do not open a public GitHub issue for security vulnerabilities.

Email security@whereas.law (placeholder until domain is configured) with:

  • A description of the vulnerability
  • Steps to reproduce
  • Affected versions
  • Your assessment of severity

We will acknowledge receipt within 72 hours and provide a status update within 7 days.

Supported versions

Whereas is pre-v0.1. There are no supported versions yet. Once v0.1 ships, we will document a support policy here.

Threat model (informational, will evolve)

Whereas is designed to run inside a tenant's infrastructure. The threat model assumes:

  • The deployment is operated by the tenant (a law firm or in-house legal team).
  • End users are authenticated employees of the tenant.
  • The LLM provider is either local (Ollama) or a remote provider configured by the tenant.
  • Documents may contain sensitive client information.

What Whereas defends against:

  • Cross-organization data leakage (multi-tenant isolation, even though most deployments are single-tenant).
  • Hallucinated metadata leaking into critical workflows (span validation).
  • Unauthorized document access (permission checks on every read).

What Whereas does not defend against:

  • A malicious tenant administrator.
  • A compromised LLM provider (mitigation: keep extraction local).
  • Side-channel attacks on the host OS.

If you find a gap in this threat model, please report it the same way as a vulnerability.

There aren't any published security advisories