Writing

Notes on AI security, agent red teaming, and behavioral detection. Working out loud, failed attempts included.

• Ten thousand agents, and none of them can tell a command from an attack 2026-06-04
• How I'd build an autonomous red team for AI agents 2026-06-04
• MCP-Poison-Bench: a de-circularized tool-poisoning benchmark 2026-06-03
• PARALLAX: behavioral threat detection from metadata alone 2026-04-09
• KESTREL: cloud-workload anomaly detection that met the data where it lived 2026-04-08
• CTF records: prompt-injection practice log 2026-04-01