74. That's how many serious privacy incidents Meta says its LLM-written catching tests flagged at the diff stage in the first half of 2026.
74. That's how many serious privacy incidents Meta says its LLM-written catching tests flagged at the diff stage in the first half of 2026. My read: Meta is using AI to police its own AI-speed code, and from a $META investor's seat that's quiet cost discipline on privacy risk.
- Meta's October 29 engineering blog says Catching JiTTests is now part of its internal infrastructure, moving past the research-prototype stage. The system writes executable tests meant to expose behavioral changes in a code change before it lands.
- Meta reports 589 engineers engaged with it, a 60% true-positive rate, and a 20% drop in privacy-SEV occurrence. Those are company numbers, and I'd hold off crediting the whole decline to one tool.
- Privacy is expensive for Meta. Italy just ordered it to pay $887M over Facebook's photo access practices, so catching bugs before code ships is a balance-sheet issue as much as an engineering one.
- No public or open-source release is documented. Whether Meta opens any of it up is the thing I'd watch.
The stock closed up 2.90% at $641.69 last session and sat 0.48% lower pre-market on Nov 16, so nobody is trading this. I think it's the kind of boring internal fix that compounds, though I could be wrong if those 60% hit rates don't hold at scale.
Sources:
MarkTechPost: https://www.marktechpost.com/2026/11/16/meta-jittests-llm-catching-tests-privacy-bugs/
HotAI News: https://hotai.news/en/news/2026-11-16/
Not financial advice. For informational and educational purposes only.