Scott Sloan·Oct 25Three Rules That Don’t Bend: Keeping an AI SOC Defensible as the Models Get BetterModels will keep improving. These three principles shouldn’t change when they do.
Scott Sloan·Oct 23The First Ninety Days of an AI Agentic SOC: A Week-by-Week PlanThree thirty-day blocks, each with a clear definition of done, turning the whole playbook into a schedule you can put in front of your team…
Scott Sloan·Oct 21Climb the Rollout Ladder: Taking an AI Triage Agent From Shadow Mode to Steady StateFour rungs, each with entry criteria, warning signs and a way back down. Don’t skip one, and be willing to stay on a rung for months.
Scott Sloan·Oct 19Measure the Two Errors Separately: The Metrics That Matter for an AI Triage AgentAccuracy is the wrong number. Here are the six measures that tell you whether an agent is safe to rely on, and how to report them honestly.
Scott Sloan·Oct 17Making an AI Triage Agent Auditable: Frameworks to Map To, and What to LogYour auditors will ask how the agent is governed and why it made a particular call. Here’s how to have both answers ready before they ask.
Scott Sloan·Oct 15Break the Lethal Trifecta: The One Capability Your AI Triage Agent Must Never HaveA SOC agent inherently holds private data and reads untrusted content. The whole security design comes down to whether it can also talk to…
Scott Sloan·Oct 13Prompt Injection Hides in Your Alert Data: Where It Arrives, and Six Defenses Ranked
Scott Sloan·Oct 11The AI Triage Pipeline: Why the Model Never Acts AloneA seven-step architecture for an alert triage agent where plain code does most of the work, and the model does only the part a model is…
Scott Sloan·Oct 9No Golden Set? Start Here: Documenting SOC Cases So Auditors, Analysts and AI Can Learn From ThemMost SOCs don’t have 300 well-documented closed cases to test an AI triage agent against. Here’s why that gap matters well beyond AI, what…A response icon1A response icon1
Scott Sloan·Oct 7Measure Before You Build: Golden Sets, Evaluation Harnesses and Model Screening for an AI SOCThe first thirty days of an AI SOC project shouldn’t produce an agent. They should produce the ability to tell whether any agent is any…A response icon2A response icon2