Bring AI security to Databricks

AI models introduce risk in the form of vulnerabilities and erroneous predictions. Our platform executes hundreds of algorithmically-generated tests to identify risks and vulnerabilities throughout development and production. This process of continuous validation is initiated automatically during experimentation when a new model is registered to the Databricks MLflow model registry. To test production models, we also integrate directly into your model’s prediction logs stored in Delta Lake or DB SQL.

AI models need to account for anomalous and malicious inputs, as well as unexpected model behavior. This is especially true for generative AI since outputs are passed to users immediately. Robust Intelligence offers several controls to reduce risk. AI Firewall connects to Databricks-hosted prediction services so that users can ensure that their production models are secure. Our model registry enables users to choose from alternative, safer models when continuous validation exposes risk.

The ability to identify and quickly address new risks in production models and data is essential to keeping AI systems production-ready. As our platform exposes new risks through continuous validation, we send alerts in real time. Alerts and thresholds can be customized and managed to your specifications and integrate with your preferred tools and workflows.

Effective AI governance is nontrivial. Our platform translates continuous validation results into automatically-generated model cards. Our governance dashboards provides a high-level view of the status of all your AI models in production. This simplifies your audit readiness and compliance requirements and maps to AI regulatory frameworks.