Researchers Poke Holes in Safety Controls of ChatGPT and Other Chatbots

July 27, 2023

July 27, 2023

New York Times: In a report released yesterday, Zico Kolter of Carnegie Mellon University led a team demonstrating how anyone could circumvent A.I. safety measures and use any of the leading chatbots to generate nearly unlimited amounts of harmful information.

“There is no obvious solution,” said Kolter. “You can create as many of these attacks as you want in a short amount of time.”

Kolter is a DTI PI developing novel approaches to cybersecurity based on semidefinite programming. 

Read more here.

Pictured: Zico Kolter, right, a professor at Carnegie Mellon University, and Andy Zou, a CMU doctoral student. Photo by Marco Garcia for The New York Times.