Chrome Bugs Fixed by AI: The Adversarial Testing Case

Google fixed more Chrome bugs in June than in the past two years combined, and the credited driver was AI-assisted adversarial testing. The HN discussion pushed back on the simplistic 'AI writes code = bad' framing and made a more useful distinction: using AI to generate code blindly is risky, but using AI to check developer assumptions, find edge cases, and suggest refactors is where the actual gains are showing up.

Commenters were appropriately skeptical about whether the team was also working harder than usual, and whether the metric (bugs fixed) is the right one versus bugs introduced. But the core claim, that adversarial testing and fuzzing are where AI adds clean value without introducing new risk, held up in the thread.

This is a narrow but credible use case that is distinct from 'vibe coding' or autonomous agent coding. It fits into existing security workflows rather than replacing engineer judgment.