Evidence-led governance for AI systems and agents.

Turn an AI-system inventory into governance findings, reproducible evidence, and a decision-ready handoff. AI tools support human review and decision; consequential authority stays with accountable people.

Try the Toolkit · Connect an AI agent

The browser tools run locally in your browser. The MCP server is available as source for local installation or operator-hosted self-hosting; Global AI Governance operates no hosted MCP endpoint. Outputs are non-authorizing (authority_effect: NONE).

1. Open the Toolkit and select Run sample (a synthetic clean inventory) or Run example with gaps (a synthetic inventory that shows governance findings). No login or upload is required for the browser workflow.
2. Review the findings and download the AI Governance Decision Pack. Neither example is a real security assessment. The clean sample may pass its bounded checks; a passing result is not a deployment decision and does not authorize deployment.
3. The hosted Toolkit reflects current main. For a pinned implementation, use the v2.3.0 release and its quick start.

Other browser experiences: Containment Ledger for security-boundary review; Peace OS: Crisis Room for a fictional crisis-verification exercise.

• AI Governance MCP exposes three read-only tools: check_inventory, build_decision_pack, and validate_passport. Every result carries authority_effect: NONE and leaves the decision with a human.
  • Install and run locally (stdio, no network or token) · Client configuration · Auth · Self-hosting · Known limits
  • Remote use: on any non-loopback bind the server refuses to start unless GAG_MCP_ALLOWED_HOSTS is set to the client-facing hostname and GAG_MCP_BEARER_TOKEN is a random secret of at least 32 characters. Serve it only over HTTPS; requests send Authorization: Bearer TOKEN. OAuth is not implemented.
  • Provider connections (Anthropic Claude, OpenAI, xAI Grok): NOT VERIFIED. Protocol compatibility is tested with the official MCP Python SDK client and the MCP Inspector. ChatGPT on the web is not supported, because it cannot send a static bearer token.
  • The gag-mcp command, gag_mcp Python package, and GAG_MCP_* variables are unchanged since the repository rename.
• Machine interface · Agent discovery · Toolkit v2.3.0 contracts · Agent authority and passport schemas (v0.1.0-alpha.2)
• Repository handoff contracts · Global AI Assurance Envelope · Portfolio schema

Published tags remain controlling for released interfaces; newer merged code is not automatically a released contract. Integrations must preserve the source system's authority and evidence boundaries.

Use of these materials does not imply endorsement or certification. Do not include confidential, personal, regulated, or protected information in public repository surfaces.

Nine product and reference repositories plus two public infrastructure repositories. Labels below are the current public maturity and outcome from the machine-readable portfolio, which also carries lifecycle roles and source-of-truth boundaries.

• Global AI Governance Toolkit · Open app — v2.3.0 · working public reference toolkit. Reproducible Decision Pack, bounded sovereignty assessment, and optional assurance/accountability profiles.
• AI Governance Model Context Protocol Server (AI Governance MCP) — v0.1.0 preview · source only · no hosted endpoint. Read-only Model Context Protocol (MCP) tools wrapping the toolkit and passport validator; authority_effect NONE.
• Agentic AI Governance — v0.1.0-alpha.2 · experimental public alpha. Machine-readable authority, evidence, passport, validation, expiry, and revocation artifacts.
• Verified Vulnerability Governance — v0.1.3 · verified-closure pre-release. Evidence-linked vulnerability ownership, remediation, retesting, and closure.
• AI Cyber Resilience Framework — v0.1.1 · public defensive reference. Boundary assessment, evidence cards, and architecture-hardening backlog.
• Containment Ledger · Open app — Local-only browser instrument for domain/path/evidence marking and human disposition. Independent structural review instrument; shares vocabulary with the AI Cyber Resilience Framework; no code, data, or release dependency; authority_effect NONE.
• Peace OS: Crisis Room · Open app — v0.3.0-rc2 · published public browser review candidate · stable, human-validation, Godot, and Windows gates open. Fictional crisis-verification simulation and after-action review.
• Governed Systems Administration — Pre-alpha · preimplementation · independent semantic review required · no execution capability. Proposed human-governed administration request, review, validation, and evidence contracts.
• AI Governance Foundations — v1.1 · foundational governance model and template set. Inventory, ownership, risk, control, monitoring, and reporting foundations.
• Repository Governance Defaults — Account-level default repository governance files. Consistent security-boundary, claims-integrity, release-governance, and change-control defaults.
• Global AI Governance Profile — Public Start Here profile for the open-source portfolio. Portfolio truth registry, federated evidence graph, and cross-estate reproduction.

Repository maturity definitions explain what these labels do and do not establish. AI Governance Foundations is the original foundation; Containment Ledger is independent of the AI Cyber Resilience Framework.

Published release identities remain controlling. Merged development candidates do not silently become release claims. The portfolio manifest records these merged, unreleased candidates; none is a new published release:

• AI Cyber Resilience Framework: v0.2.0 Continuous Assurance Thread (published release remains v0.1.1).
• Peace OS: Crisis Room: Post-RC2 Portfolio Operating Disposition Reference (published release remains v0.3.0-rc2).
• Agentic AI Governance: OPA Enforcement Bridge and synthetic Agent Incident Readiness lifecycle reference implementations (published release remains v0.1.0-alpha.2).
• Verified Vulnerability Governance: synthetic crypto-migration closure reference (published release remains v0.1.3).

Govern → Authorize → Enforce → Observe → Verify → Contain → Recover → Measure → Learn.

Repositories do not all implement every lifecycle stage and do not share one maturity level. Follow each repository's published maturity and operating boundaries.

Evidence before claims. Accountable ownership before decisions. Verification before closure. AI assists; humans retain authority.

The tools produce reference evidence and decision support. Automated checks do not establish operational safety, independent real-world assurance, legal compliance, certification, or production authorization. A passing evaluation is not permission to deploy, accept risk, or take consequential action.

For operators and reviewers: Integrated remediation reference · NIST AI RMF mapping · Source policy · Change policy · Security policy.

This profile repository is licensed under the Apache License 2.0; each linked repository states its own license and release terms.