Every Language Model Has a Forgery-Resistant Signature

Researchers have discovered that language models produce outputs with unique geometric signatures that lie on high-dimensional ellipses, which can be used to identify the source model. This signature is forgery-resistant and naturally occurring, potentially enabling cryptographic-like verification of AI model outputs.

• →Every language model has a unique geometric signature based on outputs lying on high-dimensional ellipses.
• →These ellipse signatures are practically impossible to forge without direct access to model parameters.
• →The signature is naturally occurring in all language models and can be detected without access to inputs or full weights.
• →Researchers propose using these signatures for language model output verification similar to cryptographic authentication.
• →The technique currently works for small models but faces practical challenges for production-scale models.