Enterprise Data Agent Governance

Open resources for deciding when an AI agent may answer from enterprise data, when it must request clarification or defer, and what evidence an organization should retain.

This organization hosts the Hugging Face distribution of the Enterprise Data Agent Governance project. The current dataset contains 18 synthetic evaluation cases across three decision bands—answer, needs_definition, and refuse—alongside machine-readable controls, schemas, citation metadata, licensing, and a version-pinned Mnemiq evidence record.

Project resources

• Hugging Face dataset — evaluation cases and machine-readable governance resources
• Canonical GitHub repository — code, schemas, project files, releases, and version history
• Public reference site — framework overview and practitioner guidance
• GitBook knowledge base — navigable implementation guidance

Scope

The resources support governed enterprise-data-agent evaluation, refusal testing, control workshops, and reproducible implementation planning. The cases are synthetic and are not a substitute for deployment-specific testing, access controls, semantic ownership, data-quality checks, or human oversight.

Important disclosures

This project is independently maintained by Murray Newlands as a personal practitioner resource. It is not official product documentation for Mnemiq or Agentic Fabriq and does not constitute independent certification or validation of any vendor. Mnemiq is included as a publicly inspectable worked case study. Agentic Fabriq develops and maintains Mnemiq, and Murray Newlands advises Agentic Fabriq.

The framework is not legal, compliance, or security advice. Organizations must adapt its controls to their own risk, data, jurisdiction, and operating environment.