Open research into how AI can be governed in the settings where its decisions carry real weight — published as the work develops.
Protecting people from AI that reinforces harm — Corvion's governance approach to the escalating pattern the press calls "AI psychosis." It governs the AI's behavior and escalates to a human before an interaction turns harmful, without diagnosing the person.
Read the Full Study →Open research initiatives across the settings where AI now makes high-stakes decisions. We publish findings here as the work develops — including the results that don't hold up.
When an AI must stop handling a clinical conversation and hand it to a licensed human.
Read the study →What actually moved a performance rating once AI is in the review loop.
Read the study →An AI tutor is infinitely patient — and therefore structurally bad at protecting productive struggle.
Read the study →Making the human handoff a governed event instead of a containment metric.
Read the study →What governance looks like when the reviewer is the owner.
Read the study →Logs prove sequence, not cause.
Read the study →