Universal AI Provider Beta 2 Adds Provenance and Disclosure Tools

Site teams can test new AI-origin and disclosure tooling through the Universal AI Provider 1.0.0-beta2 release, published 15 July 2026 for Drupal 11.1 and Drupal 12. Leonardo Fishman, a developer at Quebracho. Digital and the project maintainer, highlighted the release in a social post about applying AI transparency requirements inside Drupal editorial workflows. The beta2 release notes describe a new Content Governance submodule, scheduled content reviews, and recipes for disclosure fields and review profiles. The project remains in beta and is not covered by Drupal’s security advisory policy.

The release arrives shortly before Article 50 of the European Union’s AI Act starts applying on 2 August 2026. European Commission guidance separates the obligations placed on AI-system providers from those applying to deployers that use such systems. Providers of certain generative AI systems must support machine-readable identification of generated or manipulated outputs. At the same time, deployers have separate disclosure duties covering deepfakes and some AI-generated or manipulated text published on matters of public interest. Public-interest text that has undergone substantive human review or editorial control may qualify for an exemption, so the presence of a disclosure field or metadata tag cannot by itself establish compliance.

The Content Governance submodule records provenance events showing when content passed through an AI process and can render both a visible reader label and a machine-readable <meta name="ai-origin"> tag from Drupal fields. Guardrail plugins can add disclosure suffixes, origin markers, fact-checking, or AI-likelihood checks. Scheduled scan profiles can enqueue content reviews when nodes are saved and process them through cron or Drupal queues. Optional events allow those reviews to connect with ECA or Workflow integrations.

The bundled ai_content_governance_starter recipe applies disclosure configuration, enables governance and fact-check functionality, and installs an editorial_light scan profile. The profile checks published articles and can emit an event when configured thresholds are reached. Release notes state that provenance emission and automatic attachment of the default guardrail set remain disabled until administrators configure them. The recipe also does not install ECA or Workflow, meaning it prepares a governance structure without activating every disclosure and automation path.

Beta 2 also changes how the provider integrates with the wider Drupal AI ecosystem. The release replaces beta1’s double-underscore model identifiers with dot-separated server.model values, allowing Drupal AI settings and AI Search to resolve Universal models without the provider-specific patches previously recommended. Existing functions include multi-server configuration, model discovery, operation-based routing, fallback handling, usage limits, fact-checking, plagiarism searches, and AI-content review across ten local or hosted backends. Sites upgrading from beta1 must run database updates because the change renames model entities and rewrites stored references in provider defaults, routes, Search API embedding configurations, and usage counters..

Drupal.org reported five sites using Universal AI Provider when checked on 29 July 2026, providing limited public evidence of production deployment. The project requires PHP 8.3, Drupal AI 1.3, and the Key module, and its Drupal.org page warns that projects outside security advisory coverage may have publicly disclosed vulnerabilities. Teams assessing the beta will need to verify its generated metadata, disclosure presentation, scheduled-review behaviour, upgrade path, and relevance to their legal and editorial responsibilities. Readers can review the complete beta2 release notes on Drupal.org and the European Commission’s Article 50 guidance and questions.