Definition
Plain language
Hiding an invisible signature inside AI-generated text so it can later be identified as machine-made.
As stated in the literature
Techniques that bias a language model's token sampling according to a secret key so generated text carries a statistically detectable, key-verifiable signal while reading normally; closely related to steganographic channels over model output.
Why it matters: It offers a way to tell AI-generated text from human writing, which helps trace misuse while the text still reads perfectly naturally.
For example, a chatbot can subtly favor certain word choices as it writes, leaving a hidden pattern that a checker with the secret key can later spot to confirm the text was machine-made.
Heard on the show
“… tap-to-define, with links to the related papers grouped by theme, from the Prisoners' Problem to the watermarking work this builds on. …”Episode 184 — An AI Built an Undetectable Secret Channel, And Another AI Couldn't Find It