Meta has moved its Just-in-Time Tests (JiTTests) system from a research prototype into production infrastructure, according to a company research paper presented at FSE 2026. The system leverages LLMs to generate executable unit tests specifically targeted at risky code changes, providing concrete behavioral proof of bugs before changes are merged into the main codebase.

Unlike traditional hardening tests that guard against future regressions, JiTTests operate on the diff under review using diff-aware mutation and intent-aware workflows. In the first half of 2026, Meta reported that the system caught 74 privacy-related SEVs (site event vulnerabilities) at the diff review stage. Comparative analysis showed diff-aware generation produced 4x more candidate catches than standard hardening methods.

Meta operates the system with a human-on-the-loop paradigm where engineers set policy thresholds rather than reviewing every test script. While Meta has not released open-source code or APIs for JiTTests, the architecture offers a reference pattern for enterprise engineering teams looking to automate test-driven diff verification.

Why it matters

  • Engineering teams can adopt diff-aware mutation and LLM test generation to catch logic and privacy bugs before code reaches main branches.

  • Human-on-the-loop architectures enable automated code verification at enterprise scale without overwhelming developers with manually reviewing AI outputs.

Source: marktechpost.com