<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Ai · Angus Strom</title><link>https://angusstrom.com/tags/ai/</link><description>Notes on information security, written slowly. AI security, enterprise security and purple teaming.</description><language>en-AU</language><managingEditor>hello@angusstrom.com (Angus Strom)</managingEditor><lastBuildDate>Sun, 18 Oct 2026 10:52:32 +1100</lastBuildDate><atom:link href="https://angusstrom.com/tags/ai/index.xml" rel="self" type="application/rss+xml"/><item><title>Flipping the Coin: a DEFCON workshop, and why you should build one</title><link>https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/</link><pubDate>Sun, 18 Oct 2026 10:00:00 +1100</pubDate><author>hello@angusstrom.com (Angus Strom)</author><guid>https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/</guid><description>&lt;p&gt;&lt;em&gt;What it took to build our DEFCON workshop, Flipping the Coin: endless hours of lab infrastructure, what I&amp;rsquo;d change today, and why you should build one.&lt;/em&gt;&lt;/p&gt;&lt;p&gt;At DEFCON 31 and DEFCON 32 fellow hacker 5ud0ch0p and I ran the workshops &amp;ldquo;Flipping the Coin: Red and Blue Teaming in Windows Environments&amp;rdquo;. These workshops represented a core ideal in our beliefs that Offensive and Defensive operations work in sync, and that the idea of having your career siloed to one is detrimental to one&amp;rsquo;s own growth.&lt;/p&gt;
&lt;p&gt;To pull this off, we engineered a custom Packer/Ansible/Terraform stack that scaled out to over 700 EC2 instances so every single student could have their own dedicated network. Getting there cost us roughly 150 hours of our own time just building and troubleshooting the infrastructure.&lt;/p&gt;
&lt;p&gt;Which begs the question: was it worth it?&lt;/p&gt;
&lt;p&gt;Yes. But we would do it differently.&lt;/p&gt;
&lt;figure class="fig"&gt;
&lt;label class="swap has-orig"&gt;
&lt;input type="checkbox" class="swap-toggle" aria-label="Show the original image"&gt;
&lt;span class="swap-media"&gt;
&lt;img class="dither" src="https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/presenters.dither.png" alt="Two presenters in Australian bush hats either side of an inflatable kangaroo, in front of the Flipping the Coin title slide." width="960" height="723" loading="lazy" decoding="async"&gt;
&lt;img class="orig" src="https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/presenters.orig.webp" alt="" aria-hidden="true" width="960" height="723" loading="lazy" decoding="async"&gt;
&lt;/span&gt;
&lt;span class="hint" aria-hidden="true"&gt;&lt;span class="h-hover"&gt;hover&lt;/span&gt;&lt;span class="h-tap"&gt;tap&lt;/span&gt;&lt;span class="h-back"&gt;back&lt;/span&gt;&lt;/span&gt;
&lt;/label&gt;
&lt;figcaption&gt;Two true blue soldiers ready to present in Vegas. &lt;span class="sc"&gt;Hover or tap to see the original.&lt;/span&gt;&lt;/figcaption&gt;
&lt;/figure&gt;
&lt;h2 id="why-run-a-workshop-at-all"&gt;Why run a workshop at all&lt;a class="anchor" href="#why-run-a-workshop-at-all" aria-label="Link to this section"&gt;#&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;The obvious reason is that hacking is a collaborative community born out of people teaching each other techniques on the hacking forums of old. DEFCON Workshops offer a way to give back to this history of free knowledge sharing, as the workshops themselves are a &amp;lsquo;free&amp;rsquo; component of DEFCON, although admittedly nowadays it can be very difficult obtaining tickets.&lt;/p&gt;
&lt;p&gt;If you yourself want to see changes in how people view and think about the industry, be it broadly or narrowly - a workshop is a wonderful way to demonstrate your point. For us we wanted students to come away with an understanding that offence and defence are just two sides of the coin. Too often are people purely defined as being &amp;lsquo;red&amp;rsquo; or &amp;lsquo;blue&amp;rsquo; only, whereas in reality a strong hacker shifts between the two and understands both sides of this coin.&lt;/p&gt;
&lt;p&gt;Also for yourself, teaching is one of the ways to force mastery of a subject. When you’re deep in teaching and a student asks a curve ball question, or when a step doesn’t work for a student due to their setup - you can&amp;rsquo;t hand wave these moments away, you need to be able to answer them on the fly.&lt;/p&gt;
&lt;figure class="fig"&gt;
&lt;label class="swap has-orig"&gt;
&lt;input type="checkbox" class="swap-toggle" aria-label="Show the original image"&gt;
&lt;span class="swap-media"&gt;
&lt;img class="dither" src="https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/room.dither.png" alt="A hotel function room full of attendees at laptops, facing a screen that reads &amp;#39;why is this noisy?&amp;#39;" width="960" height="723" loading="lazy" decoding="async"&gt;
&lt;img class="orig" src="https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/room.orig.webp" alt="" aria-hidden="true" width="960" height="723" loading="lazy" decoding="async"&gt;
&lt;/span&gt;
&lt;span class="hint" aria-hidden="true"&gt;&lt;span class="h-hover"&gt;hover&lt;/span&gt;&lt;span class="h-tap"&gt;tap&lt;/span&gt;&lt;span class="h-back"&gt;back&lt;/span&gt;&lt;/span&gt;
&lt;/label&gt;
&lt;figcaption&gt;A full room of hackers working through the labs. &lt;span class="sc"&gt;Hover or tap to see the original.&lt;/span&gt;&lt;/figcaption&gt;
&lt;/figure&gt;
&lt;h2 id="what-the-infrastructure-actually-was"&gt;What the infrastructure actually was&lt;a class="anchor" href="#what-the-infrastructure-actually-was" aria-label="Link to this section"&gt;#&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;Credit where credit is fully due, we based a lot of our lab infrastructure design and architecture off of XPN&amp;rsquo;s (Adam Chester&amp;rsquo;s) &lt;a href="https://blog.xpnsec.com/designing-the-adversary-simulation-lab/" rel="noopener noreferrer"&gt;Adversary Simulation Lab&lt;/a&gt; which helped us significantly cut down our development time on the labs. Despite this though it still took us a significant amount of time to fully test out and scale the infrastructure to a stable level. If I was guessing at least 150 hours of work went into the infrastructure.&lt;/p&gt;
&lt;p&gt;Our infrastructure expanded beyond the Terraform/SSM stack outlined in that post. For our needs in the lab we used Packer to bake a golden Windows AMI that had WinRM enabled so that we could then also use Ansible to configure all of the infrastructure. We found this to be more stable for our particular lab configuration and additionally we introduced a number of Linux hosts for logging, attackboxes, etc. We also hit several issues configuring vulnerabilities through SSM, that just didn&amp;rsquo;t occur with Ansible. That being said though, we did continue to hit errors also with Ansible around its usage of python interpreters that required some extensive nights googling. Your mileage may vary, but it&amp;rsquo;s good to be aware that both options exist.&lt;/p&gt;
&lt;p&gt;Once the infrastructure was live and configured we used AWS Snapshots to capture AMIs of the fully configured environment. With this it was as simple as making a python script to wrap the deployment process, and then another python script to launch a scaled lab of however many labs we needed using those AMIs, after which only a minor rerun of ansible was required to regenerate per-lab VPN profiles and adjust some minor configurations to the specific lab.&lt;/p&gt;
&lt;p&gt;We’ve open-sourced our lab below, warts and all. It’s what we built in reality on a limited time budget, not a perfectly polished reference implementation. It&amp;rsquo;s also a sanitised version, with some files such as the VPN profiles and some TF files removed, so it may be in a slightly broken state. It&amp;rsquo;s provided as is, and is better treated as reference material than something to deploy.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a href="https://github.com/0x10f2c/flipping-the-coin-public" rel="noopener noreferrer"&gt;Flipping the Coin Public&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;I wouldn’t recommend using the repo for much more than a reference, or as context for your AI tool. While it did work, it was far from polished. It fit our specific purposes well enough, and there were many improvements still to be made. Read on to see how I would approach building this nowadays.&lt;/p&gt;
&lt;figure class="fig"&gt;
&lt;label class="swap has-orig"&gt;
&lt;input type="checkbox" class="swap-toggle" aria-label="Show the original image"&gt;
&lt;span class="swap-media"&gt;
&lt;img class="dither" src="https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/infra-diagram.dither.png" alt="Lab network diagram: you connect over VPN to a Kali attack box with a SIEM beside it, into a target network of four Windows targets and a domain controller." width="770" height="687" loading="lazy" decoding="async"&gt;
&lt;img class="orig" src="https://angusstrom.com/posts/flipping-the-coin-defcon-workshop/infra-diagram.orig.webp" alt="" aria-hidden="true" width="770" height="687" loading="lazy" decoding="async"&gt;
&lt;/span&gt;
&lt;span class="hint" aria-hidden="true"&gt;&lt;span class="h-hover"&gt;hover&lt;/span&gt;&lt;span class="h-tap"&gt;tap&lt;/span&gt;&lt;span class="h-back"&gt;back&lt;/span&gt;&lt;/span&gt;
&lt;/label&gt;
&lt;figcaption&gt;One lab: every student got their own copy of this. &lt;span class="sc"&gt;Hover or tap to see the original.&lt;/span&gt;&lt;/figcaption&gt;
&lt;/figure&gt;
&lt;h2 id="what-that-same-build-looks-like-now"&gt;What that same build looks like now&lt;a class="anchor" href="#what-that-same-build-looks-like-now" aria-label="Link to this section"&gt;#&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;Designing and developing the infrastructure easily cost us 150 hours of our personal time, as we did this workshop as individuals rather than on company time. When working a full time job, and already having plenty of personal life commitments, that is a big chunk, and it doesn&amp;rsquo;t include submitting the workshop, making the slides, the many discussions we had, and so forth.&lt;/p&gt;
&lt;p&gt;This is a high barrier of entry for individuals who want to create similar workshops. However I don&amp;rsquo;t believe this barrier is nearly as high anymore with the rise of AI tools.&lt;/p&gt;
&lt;p&gt;As an experiment over a weekend I went back over our janky infrastructure and code, and using Claude Code with various models on my personal dime I recreated the infrastructure. In about a weekend I was able to make a more stable, refined, and less buggy infrastructure. The simple fact is these AI tools do excel at creating projects like these, within reason.&lt;/p&gt;
&lt;p&gt;There is an art to giving the right prompts. To achieve this I started out giving the model full context of the ask, giving it a set of designs and architectures that we first came up with before making the original infrastructure. In addition to this I set it to use multiple agents to research the web for similar examples, designs and implementations. If I had just given this a simple one sentence prompt, there&amp;rsquo;s no doubt a disaster of an outcome would have ensued. At the end of the day you control the output of these tools, and slop is still slop.&lt;/p&gt;
&lt;p&gt;I don&amp;rsquo;t want to go into grand detail of how to &amp;lsquo;prompt engineer&amp;rsquo; rather I just want to highlight that with these tools we could have easily cut our development time down into a weekend of designing and developing, and then maybe one other weekend of testing and refining the output. All in all that is a very significant reduction of time for a better outcome, which matters when it&amp;rsquo;s our limited free time being used, and it would have freed us to focus on other elements of the workshop.&lt;/p&gt;
&lt;p&gt;It&amp;rsquo;s an acceleration that lowers the bar to entry for making workshops needing infrastructure, and a net benefit when used correctly and with thought. Using AI tools to speed this up is wonderful because it lets us focus on what AI tools are not good at, and allows for more people with limited time to consider creating their own workshops.&lt;/p&gt;
&lt;h2 id="what-the-ai-tools-still-dont-do"&gt;What the AI tools still don&amp;rsquo;t do&lt;a class="anchor" href="#what-the-ai-tools-still-dont-do" aria-label="Link to this section"&gt;#&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;AI Tools fail at having human originality, they are in a way the slot (or slop) machines of the 21st century. You put the tokens in, you slap the big red button, and pray for the symbols to line up. This works well for some things: ask for an everyday CRUD app and you&amp;rsquo;ll get wonderful cheap output that probably matches what you wanted, but try to come up with a more original idea and then they start to tend towards failure states. You should own your idea for a workshop, and allow that idea to reflect your views and inner thoughts about security as a whole.&lt;/p&gt;
&lt;p&gt;Another thing that AI tools still continue to struggle with is slide generation, the pain of seeing 50 talks in a row having that classic AI slop slide layout of way too verbose information and generic layout boxes is starting to become something of a trend at conferences of late. Creating slides should remain uniquely human, it&amp;rsquo;s your idea you want to communicate and you should be deeply familiar with the slides you create in order to best present your workshop and ideas to your students.&lt;/p&gt;
&lt;aside class="callout"&gt;&lt;p&gt;&lt;span class="sc"&gt;&lt;strong&gt;Know your lab.&lt;/strong&gt;&lt;/span&gt; If you do use AI tools in the building of your workshop, and in particular of any supporting infrastructure, you still need to be deeply familiar with the code it outputs. The lab will break on the day in some unexpected way, this is just the fact of doing live demos. At that point when you need to resolve the issues for that student you better be well versed in how the lab configures so you can quickly troubleshoot their error and get them and the entire class back on track.&lt;/p&gt;&lt;/aside&gt;
&lt;p&gt;Fundamentally the point here is that AI tools can help you build the lab, but they won&amp;rsquo;t give you something worth teaching, and it most certainly won&amp;rsquo;t stand up and teach it. Make sure you extensively practise your delivery, and the stability of the lab before the delivery date.&lt;/p&gt;
&lt;h2 id="so-build-one"&gt;So build one&lt;a class="anchor" href="#so-build-one" aria-label="Link to this section"&gt;#&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;The bar has been lowered, the effort needed to create a quality workshop has never been so accessible. The plumbing that easily cost us 150 hours is now a weekend, and AI tools will readily help speed up development times. What they can&amp;rsquo;t carry is the idea, or the person standing at the front of the room presenting.&lt;/p&gt;
&lt;p&gt;Remember why you would do this in the first place. Workshops are a wonderful way to give back to the community that taught us, and to push the industry towards how you think it should look. For us that was showing to newcomers that offence and defence are two sides of the same coin, and that all too often people get stuck in the viewpoint of one side of that coin.&lt;/p&gt;
&lt;p&gt;Just don&amp;rsquo;t forget: you own your content. If you decide to use AI tools in creating this content, it is still your reputation and name going onto the workshop. Be thoughtful in your usage, and don&amp;rsquo;t create a slop grenade.&lt;/p&gt;
&lt;p&gt;Finally I want to say we both consider this one of the highlights of our careers, we cannot recommend enough to people to try making their own workshop and presenting at DEFCON. We hope to do it again some day.&lt;/p&gt;
&lt;p&gt;Greetz to all the DEFCON Workshop team and thanks for all their hard work.&lt;/p&gt;</description></item></channel></rss>