Cyber Agency Chief Uploaded Internal Files to ChatGPT Exploiting Security

A cybersecurity professional has come under enquiry in the United States after he uploaded sensitive government documents to the public version of ChatGPT.

Madhu Gottumukkala, the interim director of America's cyber defense agency, who is responsible for protecting the government's secrets, posted important contracting documents to a public version of ChatGPT after taking over the agency last summer. The actions of Madhu Gottumukkala triggered a series of automated security warnings designed to prevent the theft or unintended leaking of official secrets from federal networks.

In the summer of 2025, Gottumukkala uploaded internal US government materials to ChatGPT's public version, according to media reports of US authorities. The records were marked "For Official Use Only" to indicate that they were sensitive and not meant for public distribution, but they were not classified.

Gottumukkala joined CISA in May last year and claims that he got special permission from the agency’s Office of the Chief Information Officer to use ChatGPT, despite it being blocked for other DHS (Department of Homeland Security) employees. Later, cybersecurity monitoring systems identified the uploads on the AI chatbot in August, resulting in many security warnings in the first week alone.

According to officials, the chatbot was not a secure system approved by the government, but rather OpenAl's publicly accessible tool. In order to identify possible exposure of official data, the uploads set off automated security warnings within DHS networks. The US Department of Homeland Security has conducted internal reviews as a result of the incident, and discussions over the appropriate use of AI tools in delicate government settings have resurfaced.

After Gottumukkala failed a polygraph test he had pushed for, a number of CISA employees were let off last year. He told lawmakers he did not accept the evidence of that depiction and denied failing the test.