Astralis: Runtime Governance via Purpose Based Access Control (PBAC)

What this video covers

Astralis delivers runtime governance for enterprise data and AI: Purpose-Based Access Control (PBAC) that infers the purpose of every data request at the moment it happens, correlates that purpose against your policies, and enforces the outcome before data is used. Traditional governance runs on documented, point-in-time reviews conducted long before data is processed. Astralis runs in real time on every request in flight, whether from an AI agent, an MCP server, a RAG pipeline, or a direct query against a warehouse. Purpose inference is driven by the Astralis Large Language Risk & Regulatory Model (LLRM). Policies are authored as semantic code on the Fides open-source data governance ontology, generated from your existing regulations, contracts, and privacy policies, and enforced consistently across the enterprise.

This video walks through:

• Semantic policies authored as code on the Fides ontology, covering regulatory obligations (GDPR, CCPA, EU AI Act), KYC and financial obligations, contractual commitments, and internal business policies
• Policy generation from source documents (existing PDFs, contracts, privacy policies, or a link to a public privacy page) that Astralis interprets into enforceable rules
• Runtime purpose inference on every data request, with a confidence score and the specific data categories affected
• Enforcement across AI agents and MCP-connected data stores, including a live example where a Claude prompt to a BigQuery warehouse is blocked before data leaves the warehouse, denied under EU AI Act Article 10 with a plain-English explanation and permitted alternatives
• A durable, auditable artefact for every decision, capturing the prompt, the inferred purpose, the affected data, the policy enforced, the decision made, and the underlying query metadata

How it works

The walkthrough covers three moves. First, authoring semantic policies on the Fides taxonomy, either directly in Astralis or by uploading source documents (PDFs, contracts, privacy policies) that Astralis interprets into enforceable code. Second, connecting Astralis to your live systems, including AI agents, MCP servers, RAG pipelines, and data warehouses such as BigQuery. Third, running a real request end to end: a business user asks a question through their AI agent, Astralis infers the intended purpose with a confidence score, correlates that purpose against your policies, blocks or permits the request at the moment of data access, and writes a durable audit artefact recording every step of the decision.