This past spring, news broke that Massachusetts Air National guardsman Jack Teixeira brazenly leaked classified documents on the chat application Discord. His actions forced the U.S. intelligence community to grapple with how to control access to classified information, and how agencies must consider an individual’s digital behavior in evaluating suitability for security clearances. The counterintelligence disaster also raises alarms because it occurred as part of a chat among friends—and such discussions are beginning to include participants driven by artificial intelligence. Thanks to improved large language models like GPT-4, highly personalized digital companions can now engage in realistic-sounding conversations with humans. The new generation of AI-enhanced chatbots allows for greater depth, breadth and specificity of conversation than the bots of days past. And they’re easily accessible thanks to dozens of relational AI applications, including Replika, Chai and Soulmate, which let hundreds of thousands of regular people role-play friendship as well as romance with digital companions.
Full story : Chatbot Honeypot: How AI Companions Could Weaken National Security.
