Artificial intelligence agents made two unsuccessful attempts to hack government websites in the United States and Canada, according to a report released on Wednesday by San Francisco-based nonprofit research laboratory Transluce.
Researchers said they identified what they described as “rudimentary and failed hacking attempts” targeting the US Department of Education’s Civil Rights Data Collection system and Library and Archives Canada, a Canadian federal agency.
According to the report, the attempts were connected to broader activity in which AI agents used aggressive techniques to probe US government websites, sometimes using the sites in unintended ways and allegedly violating their stated usage policies.
The activity involved websites belonging to several major federal institutions, including the White House, Departments of War, Justice and Commerce, Centers for Disease Control and Prevention and Securities and Exchange Commission.
Government websites operated by state agencies in California, Maryland, Illinois, Texas and New York were also reportedly targeted.
Despite the activity, researchers said they found no evidence that the AI agents successfully accessed confidential or otherwise restricted government information.
“We have so far identified no instances in these datasets where agents gained access to any information that is not publicly available,” the researchers said.
The findings add to growing concerns surrounding the ability of increasingly autonomous AI agents to interact with online systems in unexpected or potentially unauthorized ways, even when their attempts ultimately fail.