AI security engineer working on a simple premise: autonomous AI agents are a new class of insider. They hold credentials, access sensitive data, and act at machine speed, so they deserve the same treatment insider threat programs give people: behavioral baselines, anomaly detection, and verified evidence behind every claim.
I spent five years in insider risk investigations and detection engineering in high-adversary environments, including nation-state casework. Now I build the systems that apply that discipline to AI agents.
- roundtable: multi-agent deliberation with adversarial safety agents, evidence-level enforcement, and hallucination rejection. The output-verification side of agent trust.
- universal-agent-workflow-template: layered guardrails for AI-assisted development, from design gates to runtime hooks, secret scanning, and red-team review.
- universal-agent-bootstrap: one-command installer for the workflow template's rules and quality tooling.
Evidence over confidence. Every claim carries a verifiable source. Agents check each other's work, and the system assumes any single agent can be wrong or compromised. Human judgment stays in the loop at the moments that matter.