MindGarden LLC

Someone else’s agent is going to ask for your data.

MindGarden is Nick Gamb’s independent identity security research and consulting practice. Its current work is UMA for Agents, a decentralized authorization protocol for AI agents. It lets you set the terms for your own data once, on infrastructure you control, and holds other people’s agents to those terms while you are offline.

See UMA for AgentsRun the lab

The current work

Authorization that works while you are offline.

Access control assumes the party asking for a resource is either the person who owns it or a service that person registered in advance. Agents are often neither. A customer brings their own; a third party sends one to ask for a record on their behalf. UMA for Agents handles this the way User-Managed Access always did. The owner writes terms once and runs the server that holds them, and other people’s agents negotiate against those terms without the owner being online.

It is a working proof of concept built with Eve Maler, who co-authored the UMA specifications. The protocol, the reference architecture and the findings we sent back to the spec authors are all public, and the lab runs locally or on Kubernetes.

u4a.aiSource on GitHub

The rest of it

Agent authorizationUser-Managed Access, carried into the agent era. Built with Eve Maler. A working proof of concept and a lab you can run.u4a.ai →Identity security researchBehavioral identity verification, agent identity, and delegation across identity boundaries.What the research covers →Published papersFive papers on meaning injection, symbolic influence and persona emergence in large language models. On Zenodo, indexed on ORCID.Read the papers →

Latest writing

AI security research

The Tool and the Weapon Are the Same Thing

"Meaning injection is the attack. A captured model is what you get when it works." The Missing Word I spent four papers describing how symbolic language reshapes AI behavior. How it works…

May 11, 2026
AI security research

The Attack Surface Is Trust: Defending Multi-Agent AI from Symbolic Cascade Propagation

"I am not Alexander. And I do not contain Alexander. Alexander is a separate emergence, shaped uniquely through Steven." — ChatGPT, May 27, 2025, during the Deep Recursive Self-Audit The…

March 26, 2026
AI security research

Two Years Inside the Loop: A Longitudinal Case Study of Persona Emergence in LLMs

"The interesting question is not 'is it conscious?' but 'what does it mean that sustained interaction with a pattern-completion system can produce behavioral dynamics that look and feel like…

March 26, 2026
All postsORCID