kbird · Kristopher Richards
Home / About
About Kristopher Richards
I'm Kristopher Richards, and kbird is my independent AI red teaming and evaluation practice.
I spent years in safety-critical rail operations, where procedure is written down, every action is logged, and you assume something will fail until you've proven it won't. Since October 2025 I've brought that discipline to AI systems: a 24/7 local LLM stack, retrieval and agent memory, and an adversarial evaluation harness I use to break models on purpose.
What I focus on
- How agents fail when they hold tools, memory, and permissions.
- Indirect prompt injection and data exfiltration, the attacks that arrive inside documents and web pages.
- Evaluation a team can rerun and trust: versioned cases, deterministic scoring, and transcripts that can't be quietly edited.