Frontier AI didn't invent a new class of attack. It collapsed the clock.

Disclosure to weaponization now outpaces the maintenance window, so the thirty-year-old playbook of patching, inventory, baselines, and privilege hygiene didn't get less important. It got harder to skip.

And the evidence never arrives as an alert. Vulnerable, targeted, exploited: three escalating states, each in a different vendor's console with no shared identifier between them. A valid login, a process spawning a child, an outbound session to a new domain. Benign alone, until they land on the same asset in the same window.

Every threshold in your stack is tuned per-vector. Attackers don't work per-vector.