Intent is unknowable.Behavior isn't.

Focused on AI security: what autonomous agents actually do under adversarial pressure, and how to make it measurable instead of taken on trust.

Writings

Why I stopped trying to read the attacker's mind, and what holds when you can't.

Reflexivity, the treadmill, and why strategy cancels when everyone can model everyone.

The instrument behind the essays: a sandboxed red-vs-blue range with a floor that holds.