We build Dolt, a version-controlled SQL database that implements Git semantics (branch, merge, diff, commit history) at the table level. One implementation — Nautobot, a network configuration management tool — uses this to support human oversight of AI-generated changes.

With EU AI Act Article 14 enforcement set for August 2026, we’ve been documenting how database version control aligns with the regulation’s requirements, and thought you’d find it helpful!

Article 14 Requirements

Article 14 mandates that high-risk AI systems be designed such that humans can:

• Effectively oversee the system during operation
• Decide not to use, disregard, override, or reverse AI output
• Intervene or interrupt the system

The Approach

Database branching provides a mechanism for staged AI output review. The AI writes proposed changes to an isolated branch. A human reviews the diff against production state, then explicitly merges, rejects, or modifies before any change affects the live system.

The Flow

https://preview.redd.it/v2utvji16yhg1.png?width=2174&format=png&auto=webp&s=828fae2fbc98e9edf82be820e1c50ab44c383cba

This produces an audit trail containing:

• The exact state the AI proposed
• The state the human reviewed against
• The decision made and by whom
• Timestamp of the action

Reversal is handled via CALL DOLT_REVERT('commit_hash') This = AI’s change is undone while preserving full history of the rollback itself.

I hope you find this helpful for building out systems ahead of the enforcement coming on August 2, 2026.

More detail: https://www.dolthub.com/blog/2026-02-02-eu-ai-act/