iCreatAI News
Skip to content
MarkTechPost· Michal Sutter·· 2 hr agoAI score58

Meet Meta’s JiTTests: LLM-Generated Tests Designed to Fail Before Risky Code Lands

Meet Meta’s JiTTests: LLM-Generated Tests Designed to Fail Before Risky Code Lands

AI summary

Meta has moved JiTTests, an LLM-based system that generates executable tests designed to fail on risky code changes, from research prototype into production infrastructure, and reports it caught 74 privacy SEVs at the diff stage in the first half of 2026 while privacy SEV occurrence fell 20%. The pipeline selects risky diffs overnight, generates candidate tests via dodgy-diff and intent-aware workflows, scores failures with an LLM judge ensemble plus rule-based filters, and keeps humans on the loop; the FSE 2026 paper analyzes 22,126 generated tests. Engineers report a 60% true-positive rate at full scale, up from 19.5% in the initial deployment, and the system is internal only with no code, models or API released.

Source: MarkTechPost · marktechpost.com