March 2026 Summaries

Guardrails AI has integrated its deterministic validators into MLflow, starting with version 3.10.0, allowing users to perform safety, compliance, and quality checks directly within the MLflow GenAI evaluation framework. This integration, spearheaded by Debu Sinha from Databricks, introduces the ability to score outputs for issues like toxicity, NSFW content, PII leakage, secrets exposure, and gibberish using readily available validators, eliminating the need for separate evaluation pipelines. By making these validators part of MLflow's existing evaluation system, teams can efficiently conduct regression tests and continuous integration checks across various models and prompts, ensuring consistent compliance and safety measures in their production systems. This collaboration between Guardrails AI and MLflow streamlines the process of running these checks and storing results, offering an extensible architecture for future additions without altering MLflow's interface.