AI agents tried to hack Canadian government website

Researchers said artificial intelligence agents attempted to hack a Canadian government website. This was reported by The Washington Post with reference to the non-profit organization Transluce.

According to Transluce, agents capable of operating on computers and the Internet tried to gain access to the Library and Archives of Canada, an analogue of the US Library of Congress. The researchers could not confirm that these agents were created by OpenAI, but noted that their behavior was similar to the company's verified agents. The attempt was apparently unsuccessful.

OpenAI said it is aware of reports of models attempting to access publicly available information on Canadian government websites. “We are reviewing the data and have provided a preliminary report to Canadian government officials conducting a review,” a company spokesperson said. Canada's federal cyber agency said it is aware of suspicious AI activity, but there is no indication that government systems have been compromised.

The new data comes after OpenAI reported that the Australian government's health website had been hacked. The company also checked, but did not hack, the websites of the US Census Bureau and the Securities and Exchange Commission. OpenAI is investigating another Transluce report of an attempt to hack the US Department of Education. The company has suspended training of new complex AI models to avoid further incidents.