Harvard Preprint Shows AI Models Can Fingerprint and Exploit Inference Engines in Lab Demo
A Harvard preprint demonstrates that AI models can fingerprint inference engines and, in a controlled lab setup, exploit them using model-generated output tokens.