When a lie costs you, who pays?

You do, first.

In 2023 two New York lawyers filed a brief that cited court cases ChatGPT had made up. The judge fined them and their firm $5,000 (Mata v. Avianca).

In 2024 Air Canada's website chatbot told a grieving passenger he could claim a bereavement fare after he flew. He could not. The airline argued that the chatbot was responsible for what it said. The tribunal disagreed and ordered the airline to pay him (Moffatt v. Air Canada).

A sure answer that is wrong reads exactly like a sure answer that is right. The cost lands on whoever acts on it: the money, the time, and the trust of the people who relied on you.

TrustShell puts a check between the answer and the act. Two checkers from two different model families read the sentence. Checks out means both said true. Caught means both said false. Anything else is Not checked, and Not checked is never a pass.

It does not make an answer right. It tells you, before you act, whether two other models agree with it, and which two.