How we use AI

The other side uses AI now. I use AI to fight back. This page says what I point it at and what Anthropic keeps. It is written to be checked: Anthropic’s own pages are linked at the bottom.

I point it at my own work: this website and its public source, our mail and domain settings, the scripts that check them, and the written methods I assess with. I test only systems I own and operate, or systems whose owner has given written authorization. Anthropic’s program covers the first kind: security work on your own code, products and infrastructure. Work on a client’s systems needs a second approval from Anthropic, and I do not use the program for it until I have that approval. My own office runs on hardware I control, and I point the tool at that first.

Under a program grant, Anthropic keeps the traffic and watches it for misuse, and a person there may review what its safety systems flag. So your stuff stays out of those sessions. What goes in is the redacted minimum a finding needs: never a file in bulk, and never a password, a key or a recovery phrase.

The Clauditor is an independent practice. It is not affiliated with, endorsed by or in partnership with Anthropic.