Kirk Abbott

Cloud security is a mature discipline. The place where cloud meets AI is not. Most cloud security tooling assumes a person is writing the requests: shared responsibility models, IAM patterns, network controls. Agents break that assumption. Least-privilege patterns designed for people compose badly with software that needs a broad action surface and produces unpredictable input, audit trails built for human-paced activity do not scale to agent-paced activity, and prompt injection, model exfiltration and adversarial input are now security problems that did not exist three years ago.

That gap is where I want to work: cloud architecture and security on one side, AI and ML systems on the other. They do not fit together cleanly yet, and I think people who can reason carefully about the join will be useful for a while.

Most of my engineering experience came from building web applications and picking up cloud infrastructure as I went. Pages, APIs, queries, deployments. Ship enough of those and you start asking what happens when one of them is wrong.

I am currently in the U.S. Air Force doing cyber defense operations, which is the same problem from the other side: defending systems somebody else built, usually without the context the builders had. Both views point at the same gap. Builders ship systems they cannot fully reason about, and defenders inherit systems they cannot fully see.

I also run Stratbeacon, a trading signals SaaS I built and operate on my own infrastructure: a multi-service stack with real users, real payments and real OAuth integrations. It keeps me close to the operational side of what I write about here.

Cloud architecture, security and AI/ML are what I want to work on. The writing and projects here are part of that.