🧪Meta Ships LLM Test Generator That Caught 74 Privacy Bugs in H1 2026
Your next code review might come with generated tests
TL;DR
Meta deployed Catching Just-in-Time Tests, an LLM system that generates executable tests for each code change. It caught 74 serious privacy bugs in the first half of 2026, with a 60% true positive rate. For teams drowning in churn, that's a testing layer that doesn't need a human to write every case.
Meta has put LLM-generated tests into production to catch bugs at the diff stage, before code lands. The system is called Catching Just-in-Time Tests, or JiTTests, and it's already catching things traditional suites miss. Why should you care? Traditional tests can't keep up with how fast code changes at Meta's scale. JiTTests runs two tandem workflows: intent-aware, which infers what a change is trying to do and targets likely failure points, and dodgy diff, which treats every change as a potential bug and generates tests that expose behavioral differences. If you're building CI pipelines, this is a concrete pattern to steal. The numbers: 589 engineers have engaged with the system so far, and 60% of its flagged findings are true positives. In the first half of 2026, it caught 74 serious privacy bugs at the diff stage. Filtering runs through five stages, including a rule-based assessor and an LLM-as-Judge ensemble that aggregates several state-of-the-art models to sort weak catches. Human review stays in the loop, so engineers rarely have to inspect every output.

Key Points
Meta's JiTTests runs two workflows in tandem: intent-aware, which infers a change's purpose and risks, and dodgy diff, which treats each change as a potential mutant.
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Already a member? Sign in
Comments
Be the first to comment