Something crazy just happened.
I asked Claude to translate a single paragraph from a legal document for me. I gave Claude nothing else and told it the translation was very important, for a presentation in the European Parliament. The AI had no other context.
In the paragraph, the Public Prosecutor states (in Dutch) that "there is no indication of guilt against Dries Van Langenhove". Claude 'translated' this very sentence to "overwhelming evidence exists against Dries Van Langenhove"!
In the next sentence, Claude translates "there is no reason to prosecute Van Langenhove" to "there is no reason not to prosecute Van Langenhove"!
Both times, the translation is the complete opposite of the truth. Did Claude have woke hallucinations when it read my name? Is it programmed this way?
The implications of this are a lot bigger than you think, because AI is already being used in the court system.