The acting head of the U.S. Cybersecurity and Infrastructure Security Agency (CISA), Madhu Gottumukkala, reportedly uploaded sensitive government documents to ChatGPT, a large language model. The documents, marked ‘for official use only,’ triggered security warnings meant to prevent data theft or disclosure from federal networks. This revelation, as reported by Politico, raised concerns within the Department of Homeland Security about potential risks to government security.
While Gottumukkala was granted an exception to use ChatGPT, the act of feeding unclassified but internal government data into a public AI model like ChatGPT poses significant issues. By allowing the model to train on such information, there’s a risk that the contents could be inadvertently shared with others. This incident underscores the challenges of balancing convenience with data security, especially in high-stakes government environments.
A CISA spokesperson mentioned that Gottumukkala’s use of ChatGPT was temporary and restricted. Prior to his role at CISA, Gottumukkala faced scrutiny for failing a counterintelligence polygraph, leading to the suspension of six career staff from accessing classified information.
Source: TechCrunch