Skip to content

Security: dvai-global/dvai-bridge

Security

SECURITY.md

Security Policy

DVAI-Bridge is a multi-platform local-AI SDK shipping under BSL 1.1 with commercial enforcement. Vulnerabilities in the SDK can affect downstream applications running in production, so we take reports seriously and respond on a predictable timeline.

Supported Versions

Only the latest minor release of the current major version receives security updates. Older minors / older majors are end-of-life — the upgrade path is the migration guide for each major.

Version Supported
v4.x ✅
< v4.0 ❌

When v4.1 ships, v4.0 stops receiving fixes; we don't backport across minors. If you need long-term support on an older minor, open a discussion about commercial support.

Reporting a Vulnerability

Do not file public issues for security reports. Use one of:

  1. GitHub Security Advisory (preferred) — open Security → Report a vulnerability. This routes the report privately to the maintainers and creates a tracked advisory thread.
  2. Email — info@deepvoiceai.co with "DVAI-Bridge security" in the subject so it routes correctly.

Encrypted email is welcome but not required — reach out before sending encrypted payloads so we can exchange keys.

What to include

  • A clear description of the vulnerability.
  • A minimal reproduction — code snippet, test app, or step-by-step instructions. We'll need to validate the issue before scoping a fix.
  • Affected SDKs / versions / platforms.
  • Impact assessment in your own words (RCE? Cred theft? DoS? Trust bypass? Cross-tenant data leak?).
  • Whether you've already disclosed to anyone else.

What to expect from us

Step Target
Acknowledgement that the report was received within 3 business days
Initial triage decision (in-scope, severity tier) within 7 business days
Patch landed in a supported release (low / medium / high severity) within 90 / 60 / 30 days respectively
Public advisory + CVE if applicable coordinated with reporter

We do not currently run a paid bug-bounty program. Responsible reports are credited in the advisory and CHANGELOG.md unless the reporter requests anonymity.

Scope

In-scope (we want reports on these):

  • The TypeScript core (packages/dvai-bridge-core/).
  • The license validator on any SDK (@dvai-bridge/core/license, Swift LicenseValidator, Kotlin LicenseValidator, DVAIBridge.License, Dart LicenseValidator).
  • The embedded HTTP server / OpenAI shim.
  • Distributed-inference offload + rendezvous code.
  • Native SDKs: @dvai-bridge/ios, @dvai-bridge/android, @dvai-bridge/react-native, dvai_bridge (Flutter), DVAIBridge (NuGet family), @dvai-bridge/capacitor*.
  • DVAI Hub desktop app (hub/).

Out of scope:

  • Third-party backends. Vulnerabilities in upstream llama.cpp, MediaPipe, MLX, Transformers.js, WebLLM, LiteRT, ONNX Runtime, or ML.NET should be reported to those projects directly. We'll bump the dependency once upstream ships a fix.
  • Example apps under examples/ — they're demonstrations, not production code.
  • Documentation typos — open a regular issue or PR.
  • Resource-exhaustion DoS that requires arbitrary CPU / memory. Local inference is inherently resource-intensive; "loading a 70B model OOMs my phone" is not a vulnerability.
  • Issues only reproducible against forks or custom builds.

If you're unsure whether something is in scope, send the report anyway — we'd rather decline a few than miss something real.

Disclosure

We follow coordinated disclosure. Default embargo is 90 days from when we acknowledge the report, extendable by mutual agreement. Public disclosure happens via:

  • A GitHub Security Advisory on this repo (assigns a CVE when the issue warrants one).
  • An entry in CHANGELOG.md under a ### Security heading for the release that contains the fix.
  • A dvai-global/dvai-bridge Discussions thread linking to both.

If a fix lands in a private branch ahead of the advisory, the public commit message uses a generic description until the embargo lifts.

There aren't any published security advisories