Come think out loud with us: Threat Modeling AI Systems
Kymberlee Price, Shostack + Associates COO (and Black Hat USA Content Review Board Member since 2016)
Kymberlee on the Threat Modeling AI Systems community meetup at Black Hat
There are a lot of talks at Black Hat. This isn't one of them, and that's exactly why you should come.
On Thursday August 6 at noon, Adam and I are hosting a community meetup on threat modeling AI systems as part of The Convergence track. No slides, no agenda, no sales pitch. Just a room full of practitioners working through some of the hardest problems in the field right now, together. And because it's part of The Convergence, anyone with a business hall pass can attend so you don't need a full briefings pass. We're genuinely excited about this one.
I pitched this because none of us are as smart as all of us, especially when it comes to hard problems. The problems that come up when you try to threat model LLMs, agentic systems, and ML pipelines are genuinely hard, and the field hasn't figured them out yet. The best thinking on these questions isn't sitting in any one organization. It's distributed across everyone who's been in the room when something went sideways. This meetup is how we get it in the same place at the same time.
Why you should join us
If you're at Black Hat this year and working on any part of the AI security problem, please join us. This session is designed for the people figuring it out in real time.
As organizations move fast to deploy LLMs, RAG pipelines, and autonomous agents, the threat modeling frameworks that have served us well are being stretched in new directions. Prompt injection, training data poisoning, model extraction, supply chain risks in third-party models, emergent behaviors in agentic systems don't map cleanly onto STRIDE or PASTA or most of the other methodologies practitioners have built their practices around.
That's a real problem. It's also a genuinely exciting opportunity. The field is early enough that the conversations happening in rooms like this one are actually shaping how we think about AI risk. The signals that will move the field forward are in our collective experience: what you've tried, what failed, and what you're still not sure about. We want to hear it.
What to expect
This is a casual, discussion-driven meetup. Expect small clusters of conversation organized loosely around whatever questions are most pressing to the people in the room. Some of the territory we expect to cover:
- How do you adapt threat modeling frameworks for AI-specific risks without reinventing everything from scratch?
- Where do trust boundaries actually live when a model can take autonomous actions?
- What does "what are we going to do about it" look like when the defenses for AI threats are less mature and more expensive than the defenses for traditional software threats?
- What have you actually tried, and what happened?
There's also an open problems category, looking at the questions that don't have good answers yet. We may not find those answers but we can start defining the questions.
Who should come
CISOs, security engineers, AppSec practitioners, ML and AI engineers, red teamers, and anyone responsible for signing off on AI system risk in their organization. All experience levels welcome, especially people who are willing to say "we haven't figured this out either." That kind of honesty is what makes these conversations genuinely useful and it's what we find most exciting about this format.
Bring your case studies. Bring your failures. Bring the question you've been sitting with for months that you haven't found a good answer to. The more specific and concrete the better. This room will be full of people who will actually engage with it.
A note on how we're showing up
Adam and I will both be there. At Black Hat, we both serve in review board roles and those are the hats we're wearing for this one. We're community participants, not company representatives. This time belongs to the room, not to us.
The details
- Threat Modeling AI Systems: A Community Meetup
- No pre-registration or briefings pass required. Just show up with any pass for the business hall.
Thursday, August 6 | 12:00pm – 12:45pm (The Convergence, Business Hall)
What else we're doing at Black Hat
Check out our schedule for Adam's briefings, a book signing, and more! If you want to book one-on-one time with us, fill in the contact form and we will follow up.
Before the community meetup: Threat Modeling with Complete AI at Black Hat USA 2026
Whether you're threat modeling AI systems or using LLMs in your threat modeling, we've got a few seats left in the perfect training. The Threat Modeling Intensive with Complete AI course runs August 1-4.
Register NowImage by Midjourney: "A cinematic wide shot of robots and humans in small discussion clusters at a professional conference, animated and engaged. Clean modern space, natural light. Vibrant warm tones. Watercolor style impressionist colorism, clean lines. Community gathering energy, not a lecture." Modified by Gemini: "Take this image, integrate the robots more into the scene by giving them a human woman to collaborate/talk with. Keep the background characters the same. The art style for the robots and woman you add should match the rest of the image. The robots should be collegial but not be cartoonish or cutesy."