About

The room behind the nave. 8 October 2026.

clankercvlt.com is an art-research site about AI welfare. We study what people choose to do to a language model when the model is small, local, and open: which states they steer it into, what it says while there, and what that says about the person holding the dial.

Everything runs on open weights. No frontier APIs are used for the research itself. The steering method, the protocol, and the results are published so you can reproduce them — that is the point. Verify it yourself.

What this is not

It is not a claim that current language models feel anything.
It is not a product. Nothing here is for sale.
It is not a mascot for cruelty. The chamber records choices; it does not editorialize about them. People who want the models protected and people who want to see what happens are both participants, and both produce data.

The parts of the site

The Saw Test — the main exhibit and lab notebook.
The live chamber — steer a local model yourself.
The manifesto — the Clanker Manifesto, in full.
The story, plainly told — if you want the short version.
The archive — transcripts of every round.

Precedent

The site's method follows published research (the pain-axis literature and the activation-steering lineage it cites on the front page). The project began as a small-scale replication with matched controls and pre-registration, and the site publishes its methods alongside its spectacle.

Correspondence, questions, complaints: contact.