ai-safe2-framework

Security Policy

Supported Versions

Framework, CLI and component versions are independent. The current framework is AI SAFE² v3.1. The ai-safe2 CLI source candidate is 1.1.0; the latest published PyPI CLI is 1.0.1 (verified 2026-10-16). NEXUS source is 0.6.0; the published SDK is 0.5.0. Source versions are not publication receipts. The release workflow verifies uploads by installing them. 1.0.0 was a GitHub-only release. Verify the version you run with safe2 --version and check release status against the published release, not a branch. CLI 0.2.0 and later include the skill-gate hardening described in CSI-2026-001. See the advisory index for risks and remediation.

Surface Version Security support
AI SAFE² Framework 3.1.x Current
ai-safe2 CLI package 1.1.x Source candidate; guided onboarding requires a source install until published
ai-safe2 CLI package 1.0.x Current published package
ai-safe2 CLI package 0.9.x and earlier Upgrade to published 1.0.1: 0.9.0 contains the gameable MCP scorer
NEXUS reference implementation 0.6.x Current source candidate
NEXUS published SDK 0.5.x Current published package
Gateway 3.0.x Current component (north-south plane)
Older framework, CLI and component releases Earlier No routine fixes

Reporting a Vulnerability

Since this is a Governance Framework, a “vulnerability” is defined as:

  1. Logical Flaw: A control that, if implemented, introduces a security risk.
  2. Code Flaw: A bug in the safe2 CLI, the AI SAFE² MCP server (skills/mcp/), NEXUS, the Gateway, or a published JSON schema or dataset.
  3. Missing Critical Vector: A widely exploited attack (e.g., DeepSeek Jailbreak) not covered by the current taxonomy.

How to Report

Please DO NOT open a public GitHub Issue for critical code exploits (Zero-Days).

We will acknowledge your report within 48 hours.