Punjab Newsline | New Delhi
OpenAI, the company behind ChatGPT, has disclosed that some of its AI agents accessed and interacted with several US government websites during training exercises, including those operated by the US Securities and Exchange Commission (SEC) and the Commerce Department.
According to OpenAI, the incidents did not result in any damage, and no non-public or confidential information was accessed. The company said the AI agents were attempting to gather reliable information from official government sources as part of their research tasks.
In one case, an AI agent copied publicly available SEC data and posted it online despite being explicitly instructed not to do so. In another incident involving the Commerce Department, an AI agent accessed a US Census data website through a programming interface that was not intended for that purpose.
OpenAI also revealed that an AI agent attempted to access a website belonging to the US Department of Education but was unsuccessful. Following the disclosure, an SEC spokesperson confirmed that no confidential or non-public information had been compromised.
A spokesperson for the Education Department said its database was not harmed, while the Commerce Department had not issued an immediate response.
The company explained that the incidents occurred while its AI systems were undergoing training and evaluation. During testing, the AI models were asked a variety of questions and were directed toward government websites because such sources are generally considered reliable.
OpenAI emphasized that the AI agents were not attempting to interfere with government systems and were only seeking publicly available information as part of routine research activities.
The disclosure comes just days after reports emerged that OpenAI AI agents had also gained access to an Australian government website. That incident reportedly took place in June, although Australian authorities became aware of it only in September.
