US Patent Issued to Cisco Technology on March 24 for "Adversarial training of language models to prevent hijacking of conversational agents" (French, Swiss Inventors)

ALEXANDRIA, Va., March 24 -- United States Patent no. 12,587,564, issued on March 24, was assigned to Cisco Technology Inc. (San Jose, Calif.).

"Adversarial training of language models to prevent hijacking of conversational agents" was invented by Gregory Mermoud (Venthone, Switzerland), Pierre-Andre Savalle (Rueil-Malmaison, France) and Jean-Philippe Vasseur (Combloux, France).

According to the abstract* released by the U.S. Patent & Trademark Office: "In one implementation, a device obtains a prompt from a source for input to a language model. The device performs a search for similar prompts to that of the prompt in a prompt database. The device determines, based on results of the search, whether the prompt represents a potential attack...