Observed arrival · 2026-09-17
Probefield puts AI agents through a red-team gauntlet
A self-serve tool that runs established open-source red-teaming engines against an AI agent’s endpoint, model, system prompt, tools, and memory.
Field notes
The service acts as an orchestration layer between an agent endpoint and external red-teaming engines, handling the adapter work so users can define a target, launch attacks, and inspect the resulting exchanges. The homepage claims coverage of 50+ attack categories and names six attack types, including prompt injection, data leakage, tool misuse, exfiltration, and privilege escalation. garak and promptfoo are listed as available engines; PyRIT is identified as planned.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue