Part 6 of Defense at Machine Speed: The Operating Model.
Final part. The first five changed one component each. This one assembles them: defense as a loop, not a calendar.
1/ The loop, five stations, never stopping.
Hunt: scan the whole estate with the best models, on the attacker's cadence. Verify: a finding earns its queue place by surviving an attempted exploit. Patch: AI drafts, humans judge diffs. Prove: re-attack every fix; it counts when the exploit fails. Deceive and tax: decoys charge the attacker; short-lived identity removes the cheap paths.
2/ Two structures hold it up.
A range: replicas of your own systems where the real harnesses, ARTEX and peers, run under your control. That is what DAST means now. Not simulating a human clicking. Facing the actual attacker, rehearsed.
3/ And measurement with the courage to be outcome-shaped: attack success rate, detection and containment time, fix survival under re-attack, cost per resolved incident.
Programs that report activity get funded to produce activity. Programs that report outcomes get funded to produce outcomes.
4/ The team changes shape. The loop's labor is machine labor: scanning, reproducing, drafting, retesting. Humans concentrate on judgment: which risks to accept openly, which fixes ship, how the next disclosure reorders the queue.
Smaller at the funnel. Stronger at the decisions.
5/ Nothing in this model requires a breakthrough. Models available, harnesses public, benchmarks published, receipts in the open: Glasswing's 99% unpatched, Cloudflare's 30x, an FBI FLASH alert.
The attackers made the transition. Defense gets to choose whether it makes it on purpose.
Full series on Substack.