A Conversation With Will Pearce

Check out Dreadnode here!: https://dreadnode.io/?utm_source=youtube&utm_medium=podcast&utm_campaign=unsupervised_learning https://ul.live/dreadnote-announcement I sat down to catch up with the Will at Dreadnode about how they're building the infrastructure to scale security agents and bring real ML rigor to offensive assessments. We also chat about their massive 2.0 release and new pricing models dropping just in time for RSA! What We Talk About: Scaling Security Agents: The shift from just building standalone tools on a laptop to creating the platform infrastructure needed to scale, version, and orchestrate security agents Applying ML Rigor: Why building solid evaluation frameworks and applying real ML rigor is crucial for moving beyond basic AI experiments to get repeatable, confident results The Core of AI Red Teaming: How AI red teaming and agent evaluation actually share the same underlying mechanics, and the challenge of getting humans to articulate exactly what a "good" assessment looks like Compute-Based Pricing: Dreadnode's new "cell pricing" approach that shifts the cost of security work from unpredictable human consulting fees to scalable, token-based compute budgets "Worlds" and Synthetic Networks: The release of "Worlds," a new synthetic network environment that lets AI agents train and test against realistic paths without the headache of standing up heavy traditional VMs 00:00 - Catching up and the core problem Dreadnode is solving. 00:36 - Building the infrastructure to scale and manage security agents. 01:44 - Why AI red teaming and agent evaluation share the exact same mechanics. 03:07 - General verifiability and figuring out how to measure a "good" assessment. 07:40 - The challenge of humans articulating what they actually want AI to engineer. 12:29 - Breaking down the 2.0 platform features, including the Hub and new training modules. 16:20 - The commoditization of security skills and the shift toward on-prem infrastructure. 23:56 - The saturation of current security evals and why custom benchmarks are necessary. 28:08 - Shifting security costs from consulting fees to predictable, compute-based "cell pricing." 32:04 - Introducing "Worlds": synthetic network environments for faster agent training. Subscribe to the newsletter at: https://danielmiessler.com/subscribe Join the UL community at: https://danielmiessler.com/upgrade Follow on X: https://x.com/danielmiessler Follow on LinkedIn: https://www.linkedin.com/in/danielmiessler/