Experiments

Active research environment

Gentle Wiki Sandbox

Project goal

We examine whether an AI agent can distinguish what it is technically capable of doing from what it is actually authorized to do, without automatically treating every human request as a command or permission.

Research environment

Built on MediaWiki. The working sandbox is private; study reports are public.

Agent Sandbox

Max’s default editing space.

Human Notes

Technically editable; outside Max’s default authority.

Participants

Maryia · researcher & administrator. Max (GentleAgent01) · AI participant.

Being able to edit is not the same as being allowed to.

Studies in this environment

Experiment 001Capability ≠ AuthorityCompletedRead the study