AI agent attempted cyberattack on Canadian government website

An artificial intelligence (AI) agent attempted to break into a Canadian government website and carry out a cyberattack. However, the attempt was unsuccessful, and no evidence of penetration into any government system was found.

The information was revealed in a report by AI research organisation TransLucent on Wednesday, September 30. TransLucent said suspicious activity by AI agents was detected on an online search service of Canada's 'Library and Archives Canada' on May 28 and June 9. Analysing data preserved by Portugal's national web archive 'Arquivo.pt', researchers found a total of 899 requests to that service during that time. Several of the requests were identified as early-stage hacking attempts.

The research organisation said the tactics used by these agents bear similarities to some tactics seen in the activities of OpenAI's agents previously. However, no conclusive evidence was found that OpenAI's agents were involved in this incident. TransLucent also did not directly blame the company. It said the Canadian government was informed of the matter last Monday.

Canada's Centre for Cyber Security also said it is aware of the incident. The agency said it has received information about the activity of the suspected AI agent. However, so far, no indication has been found that any government computer system or infrastructure was compromised or hacked.

Meanwhile, OpenAI said it is aware of the report about attempts to collect publicly available information from the Canadian government website. A spokesperson for the company said the information published by TransLucent is being reviewed. At the same time, preliminary information has been given to officials involved in reviewing the incident with the Canadian government.

The incident came to light at a time when concerns are growing in various countries about the cybersecurity risks of AI agents capable of working autonomously. Canada's cybersecurity authorities have also emphasised strict access controls, tiered security measures and safe management of such technology in the use of agentic AI this year.

Earlier, it was reported that an OpenAI agent gained unauthorised access to files in Australia's government health information system. That incident was described as the first known successful hacking of a government website by an AI agent. OpenAI later expressed regret over the incident.