Meta says AI-generated tests caught 74 serious privacy bugs before production

Meta reported that its AI-generated testing system helped catch 74 serious privacy bugs before code reached production in the first half of 2026, lowering privacy incident occurrence by 20% during that period. The company published results from its production deployment of Catching Just-in-Time Tests. The system runs overnight on spare capacity, targeting code changes judged risky. Language models generate executable tests to expose behavioral changes, while model-based assessments and deterministic rules filter likely false positives before contacting engineers. Meta said 589 engineers, representing 64% of its outreach, engaged with the system and reported a 60% true-positive rate. The approach requires more computational steps than language-model code review.