Research

Workshop: Attack Your Own Green Lights. Five Security Practices That Work With Any Stack, Taught With the Receipts From a Two-Person Shop That Runs on Them Every Day

Workshop: Attack Your Own Green Lights. Five Security Practices That Work With Any Stack, Taught With the Receipts From a Two-Person Shop That Runs on Them Every Day

Every security program has dashboards that say green. The ones that hold up are the ones whose builders try to break them first. I run a workshop on how to do that. It's built from the practices DugganUSA uses every day, and none of them depend on our tools. If your team, chapter or class wants it, message me on LinkedIn.

Who it's for

Security teams, platform and SRE teams, people building or deploying AI agents, and small shops that can't afford a lesson twice. You need no particular stack. Bring one real system you're responsible for.

The five practices

1. Green is a claim, not a fact. A write isn't done until it's read back. A check that couldn't run says "unknown," never "pass." Lab: pick one dashboard tile or alert you trust and trace it to the artifact it claims to measure. Most teams find at least one that has been reporting success about nothing.

2. Attack your own work before you ship it. Independent review that has to demonstrate each finding, not argue it. Lab: write three probes against a small sample service (one fetches a URL a user supplies) and watch passing tests miss real problems. This week our own new check passed 17 of 17 unit tests and still had three blockers: a parser that could freeze the server, redirects that turned it into a proxy, and a database outage that would have read as an all-clear. Four review passes found more than twenty issues, including two in code already in production. All were fixed before deploy.

3. Humans and AI sign together. Agents move fast. Irreversible steps need a named human, a signed record and a gate that checks the signature against exactly what is being shipped. Lab: design the approval gate for one agent action in your environment. This week our deploy gate refused me twice because the signed approval didn't name the exact image. Both refusals were correct.

4. Correct in public, with dates. The correction is the credibility. Lab: rewrite one past incident note as a public correction, with the original claim next to the fix.

5. Remember on purpose. Institutional memory is a control. The FAA's November 2026 outage closely repeated a 2009 failure whose fixes may never have been fully built. Lab: turn one lesson your team learned the hard way into something executable: a pre-change checklist item, an automated test or a freeze rule that outlives the person who learned it.

Format

  • Half day (four hours): the five practices with two labs, usually 1 and 2.
  • Full day: all five labs, and each team leaves with a written gate and checklist for its own system.
  • Talk (45 minutes): the practices and the receipts, for chapters, meetups and classes.

Remote or in person. Everything taught is vendor-neutral, and nothing requires buying anything from us.

What you leave with

A list of the green lights in your own environment that you have actually verified, one probe suite you wrote yourself, one approval gate designed for your own agent or deploy path, and one lesson turned into a control.

Why trust the teacher

Because every claim above has a dated public record. Our corrections are published in place. Our indicators are confirmed by read-back before we say they're in the feed. Our deploys are signed and ledgered. Read the posts, check the dates, then decide.

To book it for your team, chapter or class, message me on LinkedIn: linkedin.com/in/patrickdugganmn.


How do AI models see YOUR brand?

AIPM has audited 250+ domains. 15 seconds. Free while still in beta.

Audit your site now → aipmsec.com