About AI Sanity

Writing about AI is full of loaded, slippery language — rogue AI, loss of control, frontier models. AI Sanity takes a document and hands it back with that language surfaced: passages marked in amber, each expanded into what the phrase claims, where it comes from, and a plain-language restatement.

Annotations are written by editors and drafted by a language model, and the page doesn't distinguish them, because the distinction isn't yours to work around: everything published here has been left standing by a person who can edit or delete it. Two rules keep that honest.

• Every annotation quotes the document exactly. A model-drafted annotation whose quote can't be located in the real text is discarded rather than shown — the safeguard against a machine inventing what it is commenting on.
• Documents are immutable once annotated. The text you read is the text that was annotated; changing it means creating a new document.

The dictionary is the curated layer underneath: each entry documents a term's stated and operative meanings, its foundational texts, and its attachments to real legislation, with corrections tracked over time.

Try it on any document — or browse the annotated catalog.