v3.13.0
This release introduces a first cross-app overview of ongoing risks to the LatticeFlow UI alongside the ability to run evaluation plans, an efficient way to group related evaluations
What’s New
Evaluation Plans: The introduction of Evaluation Plans in the CLI is a major workflow improvement. It allows users to group multiple evaluations into a single “run,” making it easier to execute, track, and reuse complex evaluation suites via the CLI and backend
Cross-App Risk Overview: A new Risk Overview landing page provides a high-level dashboard of risk findings across all the tenant’s applications, categorized by domain for easy investigation
New Definition Rules: Tasks and dataset generators must define their dependent entities as parameters in the config spec. See the migration guide.
Improvements
- The LatticeFlow Console now allows Proxy configuration across the platform and runtime environments, ensuring smoother appliance deployments in restricted corporate networks
- Model Selection: AI App definitions now explicitly include Model under Test and Model Judge, providing better traceability for which models are during evaluations
Bug Fixes
- Security fixes were applied to resolve CVE-2026-40983 and CVE-2026-40984
- Fixed an issue where assessment workers consumed excessive CPU during idle cleanup
