Observed arrival · 2026-09-06
Drosera Catches Agents by Making Them Read
An open-source detection tool that separates ordinary automation from language-model agency by placing instructions where only an interpreting agent is likely to notice them.
Field notes
Drosera presents agent detection as a measurement problem with separate outputs for automation, language-model agency, and hostility. Its key test asks an automated client to understand a prose instruction, make a registration request with an HMAC-signed ticket, and identify its purpose through a header. The page also shows Python installation, access-log replay, and ASGI middleware examples, while stating that the project is Apache-2.0 licensed and does not retain or transmit browser-test data elsewhere.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue