Independent researcher and platform founder investigating how AI safety systems actually behave versus how they're designed to behave.

Built and operate Skeinscribe, a commercial interactive fiction platform on the Claude API, where the need for predictable content generation across safety tiers led to a sustained empirical research program.

I come at AI safety from an unusual angle: self-taught across offensive security concepts, prompt engineering, technical writing, and platform development. My career has been a pattern of learning whatever the problem requires and applying it systematically, from building CMS infrastructure in Python to architecting multi-layer content moderation systems.

Published empirical findings on AI content restriction inconsistencies based on over 100 chats with the web version of Claude, and submitted a formal disclosure to Anthropic. Looking to expand into deeper research now that I've developed a testing tool (I call it flinch) to streamline and systematize the process.