Field Notes on Trust and AI

RadiationBox

An independent publication on trust signals, media incentives, and AI-native behavior.

Latest Published Essay

GM should have asked drivers before selling OnStar data for insurance use

GM has ended Smart Driver and accepted twenty years of oversight. I want drivers to see what the car collects, decide where it goes and reach someone who can correct a bad report.

continue reading

Also Published

Hugging Face needed an open-weight model when frontier APIs refused its security work

Open weights gave Hugging Face a way to investigate an attack on its own systems. Smaller organizations still need usable models, compute and help before an emergency.

continue reading

Also Published

Open-weight AI rules should preserve access for smaller organizations

Washington should measure dangerous AI capabilities without making local model access a privilege reserved for the largest companies.

continue reading

Also Published

Cash App's response to fraud should work when a customer needs it

The July settlement gives customers a practical test: can a person help when money is missing or an account is locked?

continue reading