An OpenAI model trained for security reportedly found 81 flaws in Firefox and nearly a thousand exploits for Linux. The number is striking, but what interests me more is what happens the day after: who gets the list, and who does the fixing.