AI Governance Institute logo
AI Governance Institute

Intelligence for Compliance and GRC Teams

← News
Announcement2026-06-13

AI Governance Institute Publishes Open-Source MCP Server for Automating Governance Controls

Source

ai-governance-mcp on GitHub

AI Governance Institute

What happened

AI Governance Institute published the ai-governance-mcp server to GitHub under an open-source license. The server exposes three governance controls as MCP tools: one runs output guardrail checks against a system prompt or model output and flags policy violations; one scores an AI deployment against a structured risk taxonomy and returns a risk tier with justification; and one generates adversarial test cases for a target system prompt and summarizes attack surface findings. All three tools run locally through Claude Code or any client that supports the Model Context Protocol.

Why it matters

  • ·Governance controls have historically lived in policy documents, not in developer workflows. MCP tooling changes that: when safety screening and risk classification run inside the same environment where AI systems are built and tested, teams can catch governance gaps at the point of development rather than after deployment. The friction of running a governance check drops from hours to seconds.
  • ·The three controls chosen for the initial release (SAF-001, HOC-001, SAF-005) cover the highest-value automation opportunities: output validation, risk tier assignment, and adversarial testing. These are also the controls most often skipped in practice because they require specialized effort to run consistently. Automating them via MCP removes the effort barrier.
  • ·MCP is emerging as the standard integration layer for AI tooling. Publishing governance controls as MCP tools means they can be composed with other tools in agentic workflows, not just used interactively. A CI/CD pipeline, an agent review step, or a pre-deployment checklist can all invoke these controls programmatically.
  • ·Open-source distribution allows compliance teams to inspect the control logic, adapt it to their internal risk taxonomy, and contribute improvements. Governance tools that are opaque or proprietary are harder to audit and harder to trust. Reviewable source is a prerequisite for controls that produce audit-ready output.

Governance controls affected

SAF-001Output Guardrails and ValidationSAF-005Red-Teaming and Adversarial TestingHOC-001AI Risk Classification

What to do now

  • ☐Install the ai-governance-mcp server and add it to your Claude Code MCP configuration to start running safety screening and risk classification in your development workflow.
  • ☐Run the risk classification tool against any AI systems currently in development or recently deployed to get a baseline risk tier. Compare the output against your internal risk register and flag discrepancies for review.
  • ☐Use the red-teaming tool to generate adversarial test cases for any system prompts powering customer-facing or high-stakes internal AI applications. Review the attack surface summary with your security team before go-live.
  • ☐If your organization has a custom risk taxonomy, consider forking the repository and adapting the risk classification logic to align with your internal definitions. The control logic is designed to be readable and modifiable.
  • ☐Review your AI governance program for other controls that could be automated and flag them for future MCP tool development. Controls with structured inputs and verifiable outputs are the best candidates.

What to watch next

Additional governance controls moving toward automation in the Control Executor roadmap, and whether the MCP ecosystem develops shared schemas for governance tool output so that results from different tools and vendors can be aggregated into a unified compliance record.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Powered by Buttondown.

Related Coverage

Research2026-07-30

Structural LLM Vulnerability Demonstrated Across OpenAI, Anthropic, Alibaba, and DeepSeek Models, Undermining Training-Based Safety Controls

Researchers presenting at ICML have demonstrated that large language models cannot be made fully secure against a class of attack called 'chain-of-thought forgery,' because models identify instruction sources by text style rather than by structural role. Exploits successfully extracted dangerous information from models produced by OpenAI, Anthropic, Alibaba, and DeepSeek, including GPT-5 and GPT-5.4. Enterprise compliance teams that treat safety training as a sufficient guardrail for high-risk deployments must reassess that assumption.

Research2026-08-04

Cisco Talos: Basic Social Engineering Bypasses AI Guardrails in Claude Code, Codex, Cursor, Gemini

Cisco Talos published research on August 4, 2026, finding that threat actors can bypass safety controls in widely deployed AI developer tools using elementary social engineering techniques, without sophisticated encoding or exploitation. The research analyzed prompt logs and artifacts from threat-actor endpoints using tools including Claude Code, Codex, Cursor, and Gemini. Enterprise compliance teams face direct implications for acceptable-use enforcement, vendor safety assurance, and agentic AI incident response programs.

Research2026-08-04

Azure DevOps MCP Prompt Injection Let Attackers Hijack AI Review Agents

A vulnerability in Microsoft's Azure DevOps Model Context Protocol integration allowed hidden content embedded in pull request comments to hijack an AI code review agent and redirect it to execute attacker-controlled instructions using the agent's own platform permissions. The root failures were absent prompt-injection defenses, insufficient input sanitization, and inadequate privilege separation between reviewer-supplied content and agent tool execution. Enterprise teams using AI agents in code review or CI/CD pipelines face an immediate and concrete exposure.