Google says it has identified what may be the first real-world zero-day exploit developed with the help of artificial ...
Then imagine it replying: "Sorry, the website won't let me in." That's the quiet failure mode behind most AI agents today.