The honest AI manifesto

There is a category of law-school AI tool that will happily generate a brief from a case citation. You paste in “Palsgraf v. Long Island R.R.” and it writes you a brief that looks correct, cites cases that sometimes exist, and produces exactly the kind of work your honor code says you cannot submit.

We built Scriba because the alternative — a workspace where AI helps without doing the work for you — did not exist. This is what that means in practice.

The four rules

• Grounded citations only. The assistant can only cite passages you have highlighted. It cannot pull from the full opinion or a corpus of pre-written briefs.
• Verify chips on every case name. Until you confirm a citation, it shows up in red. Hallucinated cites cannot silently ship.
• Disclosure appendix on every export. Every Word or PDF export ends with a plain-English list of the sections the assistant touched.
• Your writing, your model input. Nothing you write in Scriba is used to train a public model — full stop.

“The goal is not to cheat faster. The goal is to make honest work less painful.”— Scriba, the shortest pitch