💧 From November 7, text from seven Grok models on the API carries an invisible watermark by default. Its maker says it's for the EU AI Act.
No extra characters: the model just shapes which tokens it picks.
Can a mark you can't see keep AI text honest?
t.me/the_effector
