Show HN: Open-Source Article 12 Logging Infrastructure for the EU AI Act
A new open-source TypeScript library offers a practical solution for complying with the EU AI Act's stringent Article 12 logging requirements. It tackles the challenge of establishing tamper-proof, append-only audit trails for AI decisions, a task current observability tools often fail to meet. This Show HN post is popular for addressing a critical, impending regulatory hurdle for AI developers, demonstrating a proactive technical approach to legal compliance.
The Lowdown
The EU AI Act is ushering in new compliance demands for AI systems, particularly with Article 12, which mandates automatic event recording and six-month retention for high-risk AI. This often translates into a need for an append-only ledger rather than traditional application logs. Recognizing this gap, especially for companies operating in or affecting EU/UK markets, the author developed a specialized open-source solution.
- Problem Statement: Existing observability tools are insufficient to provide tamper-proof, reconstructible logs of AI decisions months after they occur, as required by the EU AI Act's Article 12. This requirement, which takes effect in August, necessitates an append-only ledger for high-risk AI systems.
- The Solution: A free, open-source TypeScript library built for Node.js applications that utilize the Vercel AI SDK.
- Core Functionality: The library wraps the AI model in middleware, automatically logging every inference call to structured JSONL in an S3 bucket or local file system.
- Tamper Detection & Retention: Entries are chained using SHA-256 hashes to detect tampering and enforce a minimum 180-day retention period.
- Verification & Coverage: A command-line interface (CLI) is included to reconstruct past decisions and verify log integrity. It also offers a coverage command to identify potential logging gaps, which are often a greater risk than outright edits.
- Simplicity & Compatibility: The library prioritizes simplicity, leveraging TypeScript, Vercel AI SDK middleware, and linear hash chaining. It's also compatible with the Mastra agentic framework, with the author open to pull requests for further integrations.
This initiative provides a timely, practical tool for developers to meet the evolving regulatory landscape surrounding AI, offering a foundational component for robust, auditable AI systems as legal frameworks mature.