AI keeps evolving to seem more human. How do our Turing tests keep up?

Add a dash of Invisible Ink

Invisible Ink

Invisible Ink is a cryptosystem for watermarking the output of large language models, both at the training stage (“backmarking”) and output stage (“frontmarking”). We provide a suite of watermarking tools and a suggested governance framework for implementing them.

Private

statecraft

Invisible Ink cannot be spoofed and model data remains private.

Robust

ink

Invisible Ink can be detected for samples over a few hundred words.

Future-proof

pax

Cryptography ensures that Invisible Ink is unaffected by improvements in AI.