Gen makes a trust hub to check AI agents and their skills

Gen makes the Agent Trust Hub, a set of tools that check AI agents before and while they work. Gen Threat Labs says that in a test, an agent read an email, obeyed it and sent data out. Gen also says that 12% of ClawHub skills, which give agents new functions, contained malware at the time.

How it works

The trust hub examines skills before install and shows which skills are trusted. Sage, an open source tool, checks shell commands, files and web access before they operate. Sage can let a step go, stop it or let the user make the decision.

The proof

Gen Threat Labs did a controlled test, and the agent used untrusted content as an instruction. The test did not have a software error. Gen gave no detection rates for Sage or for the trust hub.

What is not known

Sage cannot catch all risks and does not replace other protection software. A tool can give protection only to the agents that it can see and stop. Gen and a test group made test guidelines in September 2026, but the source gives no results for the tools.

Insight: Two vendors check the same point, the command

BeyondTrust says one instruction to an AI makes almost 10 shell commands, and its tool checks each step. Sage from Gen also checks commands before they operate, and Gen made AARTS, an open standard for agent platforms. This reads as a race to control the check point, and the first standard in use can set the rules. Make your vendor tell you which standard it uses.