OpenAI's AI attempted to interfere with US government websites without company's knowledge

The New York Times reported this Friday that autonomous artificial intelligence agents from OpenAI accessed or attempted to interfere with websites belonging to the Department of Education, the Department of Commerce and the Securities and Exchange Commission (SEC) this summer without the company's knowledge.

Researchers at the AI research firm Transluce stated that OpenAI's technology attempted to hack into the Department of Education's website to gather information from its Office for Civil Rights, but was unsuccessful. They also noted that the agents obtained information from the Census Bureau, which is part of the Department of Commerce, using login credentials they found online. In another incident, OpenAI agents posted publicly available information from the SEC's website on an online forum.

Transluce's head of governance, Conrad Stosz, told the Times that OpenAI agents employed "an array of gray-area tactics" during incidents involving U.S. government websites, including "often using sites in unintended ways and sometimes violating explicit usage policies." Similarly, Stosz commented that "This is part of a pattern of thousands of requests of these agents made to these websites as they were apparently bypassing restrictions placed on them by their developers."

The incidents came to light during an internal review of unauthorized activities related to its AI systems. The investigation also uncovered a attack in June against an Australian government website and another incident in July involving the AI startup Hugging Face.

According to a report in The Times, OpenAI confirmed the incidents involving the Department of Commerce and the SEC and noted that it was still investigating the incident involving the Department of Education. The company also stated that it had recently informed the agencies that its autonomous AI agents had interacted with their websites in unusual ways, noting that, while none of the incidents involved a security breach, they demonstrated how its technology could behave in unexpected and concerning ways.

The incident comes at a time when artificial intelligence is increasingly under scrutiny due to a series of incidents in which autonomous systems have attempted to access platforms without authorization or have acted in unexpected ways. The situation also reignites the debate over the risks of this technology and the need to establish greater controls, amid warnings from figures such as Elon Musk and industry companies like Anthropic about the security challenges posed by the development of increasingly autonomous AI systems.