New: See CompFly govern a coding agent · 1:22

Insights on the future of Trusted Autonomy

OpenShell illustrated as a glass enclosure connected to CompFly by a continuous inspection loop labeled Observe. Test. Improve.
Integrations

Governing OpenShell Agents with CompFly

Bring OpenShell-based agents into your enterprise approach to controls, visibility, and evaluation.

October 19, 2026
Railway tracks meeting at a switch, one line curving away to the left while the other continues ahead, with fallen leaves on the ballast.
Research

Status or Stop: Testing Decision Models as Agent Controls

Testing TypeSafe AI’s JEV decision model on 500 tool-action cases, prompt-injection suites and a four-model comparison: where it helps agent controls and where it fails.

October 1, 2026
A wooden Trojan horse with a hollow body and small windows, photographed against a pale blue sky.
Research

Evaluating Prompt-Injection Controls Beyond Detection

What our AgentDojo experiments revealed about attack success, task completion, and blocking tool responses.

September 21, 2026
A stone hillside fortress of layered defensive walls and buildings climbing a steep, forested mountainside.
Thought Leadership

Controlled Agentic Autonomy: A Defense-in-Depth Imperative

Why AI Agents Need Defense-in-Depth Before They Outsmart Their Guardrails.

April 28, 2026
An aerial view of adjacent farm fields in different crops and colours meeting at hard boundaries — a patchwork of separate plots.
Analysis

The State of Agentic Security Tools in 2026

A definitive guide to the fragmented security stack: Observability vs. Guardrails vs. Runtime Control Planes.

Feb 11, 2026
A red-and-white windsock stretched fully horizontal by strong wind on a harbour mast, under a moody overcast sky with sailboats on the water below.
Open Source

Introducing Crosswind: Security Evaluation for AI Agents

An open-source platform that tests your AI agents against adversarial attacks before attackers do.

Feb 09, 2026