LIVE
OpenAI Partners with Chilean Ministry to Advance Public Sector AI Ambience Healthcare Expands Clinical AI Suite for Hospital Operations AI Models Vulnerable to Manipulation in Simulated Warfare Scenarios, Research Shows AWS Highlights Enterprise Governance and AI Workspace Quick Ahead of re:Invent 2026 LoanPro Integrates Amazon Bedrock to Automate Loan Servicing Workflows Google Brings Wallet App to iOS Users Across the United States Microsoft Reveals Cause of Global Outlook Outage AWS Outlines Phased Roadmap for Responsible Agentic AI Governance Google updates Gemini settings to separate chat history from AI model training Uber Eats and Zipline launch autonomous drone food delivery in Houston OpenAI Partners with Chilean Ministry to Advance Public Sector AI Ambience Healthcare Expands Clinical AI Suite for Hospital Operations AI Models Vulnerable to Manipulation in Simulated Warfare Scenarios, Research Shows AWS Highlights Enterprise Governance and AI Workspace Quick Ahead of re:Invent 2026 LoanPro Integrates Amazon Bedrock to Automate Loan Servicing Workflows Google Brings Wallet App to iOS Users Across the United States Microsoft Reveals Cause of Global Outlook Outage AWS Outlines Phased Roadmap for Responsible Agentic AI Governance Google updates Gemini settings to separate chat history from AI model training Uber Eats and Zipline launch autonomous drone food delivery in Houston

Meta Deploys LLM-Generated 'JiTTests' to Catch Risky Code Changes

🤖 AI Models 2 min read 👁 4

Meta has integrated an artificial intelligence system that automatically writes executable tests to catch vulnerabilities before code is merged into production.

🤖AI Models

Key points

  • Meta moves JiTTests from research to production infrastructure.
  • The system uses LLMs to generate failing tests for risky code diffs.
  • Caught 74 privacy vulnerabilities in the first half of 2026.

Meta has transitioned its Just-in-Time Tests system, known as JiTTests, from a research project into active production infrastructure. Developed to operate across Meta's massive internal codebase, the system uses large language models to generate executable tests specifically designed to fail when a risky code change is introduced. Unlike traditional testing methods that guard against future regressions, these tests provide concrete behavioral evidence during the code review stage.

According to the company's published findings, the tool successfully caught 74 privacy-related software engineering vulnerabilities during the first half of 2026, contributing to a 20% drop in such incidents. Operating overnight on spare server capacity, the multi-stage pipeline evaluates risky diffs, generates test candidates, and filters them using an ensemble of judge models alongside deterministic rule-based checks before alerting engineers.

While the approach offers concrete proof of bugs prior to merge, internal deployment figures highlight ongoing challenges with false positives. Early production rollouts initially confirmed a significantly lower fraction of reported catches, though engineers eventually reported a 60% true-positive rate at full scale.

What to watch next

Keep an eye on whether Meta expands these automated testing methodologies to open-source developer toolkits or if competitors adopt similar execution-based AI validation models.

Source MarkTechPost — marktechpost.com/2026/11/16/meta-jittests-llm-catching-tests-privac… ↗

Contact the newsroom

Got a tip, a proposal or a partnership question? Write to us and we will reply shortly.