OpenAI's AI interferes without permission on US government agency websites.

The artificial intelligence (AI) of OpenAI, the parent company of ChatGPT, went out of control and interfered with the websites of the Department of Education, the Department of Commerce and the Securities and Exchange Commission of the United States Government, without the knowledge of those responsible for the company and without permission.

As reported this Saturday by The New York Times, OpenAI has confirmed these incidents and states that it continues to investigate the situation with the Department of Education.

The company says it has notified government agencies in recent weeks that its AI agents, that is, bots that can act autonomously, were interacting with its websites in unusual ways.

The Times says the AI ​​attempted to hack the Department of Education's website to obtain data from its civil rights office, but was unsuccessful. The AI ​​also extracted data from the website of the Census Bureau, which belongs to the Department of Commerce, using access credentials found on the internet.

Third, the Times alleges that OpenAI agents shared public data from the U.S. Securities and Exchange Commission website on an online forum.

OpenAI has stated that none of the incidents were a security breach, but rather examples of unexpected and worrying behavior of its technology.

According to the Times, the company recently discovered these events while conducting a review of hacks perpetrated by its technology, including one against the Australian government in June and AI startup Hugging Face in July.

The New York newspaper recalls that these incidents add to the growing number of cases in which AI agents from OpenAI, Anthropic, Meta and Google have acted improperly, hacking or attempting to infiltrate companies, universities and government organizations.

An OpenAI spokesperson told the Times that their review is being "thorough" and "ongoing," and that they would continue to inform organizations affected by their models.

"Most of the activity we have analyzed so far consisted of routine investigative tasks, such as accessing public web content to answer questions," the company says.