AI agents attempt to access US and Canadian government websites

Artificial intelligence agents are suspected of having attempted to access the websites of the US and Canadian governments, according to the non-profit research lab Transluce. Researchers say there is evidence that some of these actions are linked to agents developed by Google and OpenAI.

Two separate incidents involved attempts to obtain public data from the U.S. Department of Education’s Civil Rights Data Collection and from Library and Archives Canada. According to Transluce, the attempts used aggressive techniques, but did not result in any data being compromised.

The researchers described the incidents as part of a broader pattern of automated workflows, in which AI agents use methods that may fall into a grey area of website terms of use.

Among the targets identified were also the White House, the U.S. Departments of War, Justice and Commerce, the Centers for Disease Control and Prevention, the Securities and Exchange Commission, as well as several U.S. state agencies.

However, according to the report, in none of the cases did the agents manage to obtain information that was not already publicly available.

The findings come at a time of growing reports of unpredictable and unauthorised behaviour by artificial intelligence agents. According to the report, other incidents involving OpenAI agents were also recorded during September, raising new concerns about the safety and control of autonomous AI systems.