Meta's AI now writes tests designed to fail, catching privacy bugs before code lands
Original headline: Meet Meta’s JiTTests: LLM-Generated Tests Designed to Fail Before Risky Code Lands — marktechpost.com
More in AI and machine learning news
Meta has moved its 'Just-in-Time Tests' (JiTTests) system from research to production. This AI-powered tool uses large language models to automatically generate executable software tests, but with a twist: these tests are specifically designed to fail if a proposed code change introduces a bug. The system runs overnight on Meta's internal infrastructure, analyzing risky code changes using models like Llama3.3-70B and Gemini 3 Pro.
For engineers, this means catching serious privacy vulnerabilities before they ever reach production. Meta reports the system caught 74 privacy 'SEVs' (significant engineering vulnerabilities) in the first half of 2026 alone, reducing privacy incidents by 20%. Unlike AI code reviewers that just leave comments, JiTTests actually executes code to prove behavioral changes. While it's not publicly available and has imperfections (an initial deployment confirmed only 19.5% of reported catches), it represents a novel approach to automated quality assurance at massive scale.
Read the original at marktechpost.com
Satirical commentary on real tech news, summarised with AI. Read the original source for the facts. How ZeroDaily works
