AI and Accuracy
AI can create enormous leverage, but confident output and polished citations are not substitutes for verification.
Security engineering · Agentic AI safety
Useful systems.
Honest boundaries.
Notes and small tools exploring security engineering, agentic systems, and enforceable boundaries.
agent
tool
policy
world
boundary status unresolved by design
Current question
What changes when software can pursue a goal, choose a tool, and cross a boundary?
Profile / 01
Security assumptions meet systems that act.
This site examines practical cybersecurity: architecture, operations, hardening, and the distance between a control on paper and a control that survives contact with reality.
The current focus is agentic AI: independently enforced containment, evaluations that test consequences rather than manners, and workflows that keep humans meaningfully in control.
- Focus
- Security engineering
Agentic AI safety - Methods
- Threat modeling
Control evaluation
Adversarial testing - Approach
- Evidence-led
Boundary-focused
Selected writing / 02
Notes from the edge.
Not Everything Should Use AI
Generative AI is useful when imperfect output is acceptable and constrained. Most software still needs deterministic control paths.
Omarchy
A security engineer's first impressions of a polished Linux desktop with unusually good agent integration and defaults that demand isolation.