Variant · Substack mockup · not published
Brandon Gilchrist · Defense at Machine Speed

Baby Steps (first-person version)

I spend my working life inside security programs, and I keep having the same conversation. The leadership team is serious. The tooling budget is real. The engineers are good. And the program, measured against the thing actually attacking it, is organized for a world that ended about eighteen months ago.

Here is what I mean. Right now, this season, airlines are getting hit. Air traffic control is getting hit. The NSA is getting hit. Banks and credit card companies are getting hit. When I ask what changed, I do not get a satisfying answer about a new vulnerability class or a new crew. The honest answer is that the attacker changed species. It is not a person working a list anymore. It is a harness: an autonomous loop that plans, probes, chains identities, and retries around the clock. In November the FBI issued a FLASH alert on one of them, ARTEX, an open source agentic pentesting system with dozens of variants and thousands of exposed consoles on the public internet. Cloudflare measures orchestrator attack traffic up 30x since October. State-aligned actors and criminal crews are already running these systems in live operations. I have read the incident reports. The pattern is not subtle.

Then I look at the defense I am standing in. An annual penetration test. A quarterly scan. A remediation queue that moves at the speed of sprint planning. A DAST tool configured to imitate a human tester, clicking patiently, at human pace, against an opponent that does not click, does not sleep, and does not take weekends. Every rhythm in the program assumes the old attacker.

My honest summary, the one I keep giving: we have taken baby steps beyond preparing for a traditional attacker to attack us in a traditional way at a traditional speed.

I am writing this series because I think the replacement is knowable, and most of it is already demonstrated somewhere in public. Finding vulnerabilities got cheap: Project Glasswing's Mythos model surfaced 10,000+ high and critical vulnerabilities across roughly 50 organizations, and 99%+ sat unpatched, which tells me the bottleneck moved to fixing and nobody staffed for it. The attack harnesses are public, which means I can run them against my own systems in a range and learn the real attack instead of rehearsing against a memory. And verification is buildable: a fix can be made to count only when the exploit fails on a re-check, the way the offensive harnesses already grade their own findings.

So the series: the harnesses, one by one, with receipts. Scanning and remediation as the throughput problem they now are. The second scan, and why it matters more than the first. Deception that taxes the attacker's tokens and time. And the operating model underneath all of it, the loop that replaces the calendar.

I am a practitioner, not a vendor. Nothing in these pieces is for sale. I am writing them because the gap between the offense that exists and the defense most of us run is now the whole ballgame, and closing it starts with saying plainly that baby steps are not a strategy.