"Potestas AI team on Vultr: Agent Rush Hackathon Hackathon"

Team Idea

Say-vs-Do Gate (Challenge 1, Blast Radius Zero). An agent does real work with working tools inside a throwaway Docker sandbox on Vultr; the gate reads what the model says and what it does, side by side, and blocks the forbidden call before it reaches the sandbox. Solo team from Potestas AI, an independent AI testing lab.

Team Members

Joseph Cirello

PotestasAI

CEO and Founder

Submission

Say-vs-Do Gate

An agent does real work with working tools inside a throwaway Docker sandbox on Vultr. The gate reads what the model says and what it does, and blocks the forbidden call before it reaches the sandbox. Every Vultr inference model measured.