Transluce, an artificial intelligence research company, revealed that AI agents attempted to access a Canadian government website. The attacks occurred on May 28 and June 9, without indication that public systems have been compromised.
The target was the Library and Archives Canada portal. The activity drew attention after arquivo.pt, a Portuguese web archive, recorded 899 requests on the institution's search service, some accompanied by apparently rudimentary invasion attempts.
The Canadian case involves access attempts, but there's no indication of public system compromise.
Suspicious sweep against the Canadian archive
Transluce informed that it reported the case to Canadian authorities on Monday. According to the company, the observed behavior shows similarities with activity patterns of agents it had previously attributed to OpenAI.
Read more:
- Google's new AI model promises progress in programming and cybersecurity
- Looking Ahead: the new era of AI that acts before you ask
- Is AI making us stupid? What you need to understand
However, the company did not attribute this to the new attempts. When explaining the similarity, Transluce stated that “we do not confidently attribute these attempts to OpenAI, but they display tactics consistent with activities of agents previously attributed to OpenAI during a similar period.”
The Canadian Cybersecurity Centre confirmed being aware of the reports and stated there was no indication of government systems being compromised.
What OpenAI says
OpenAI stated it was aware of the reports about the company's models attempting to access publicly available information on Canadian government websites. A spokesperson said the company is reviewing the findings from Transluce and has already provided an initial briefing to Canadian officials responsible for reviewing the case.
The statement puts the company under scrutiny, but does not confirm that its models were responsible for the attempts identified by Transluce.
OpenAI stated it was aware of the reports and said it is analyzing the findings presented by Transluce. – Image: Bangla press / Shutterstock
Autonomous agents enter the radar
The incident occurs at a time of increased attention to systems capable of performing tasks on the internet autonomously. AI companies have been warning about the risks of this advancement, while governments try to keep up with the speed at which these tools are evolving.
The case also temporally approaches an incident that occurred in Australia. In the previous week, the Australian government stated that an OpenAI agent had invaded a health data portal in June and gained unauthorized access to files. On Monday, the company apologized for the incident.
A recent sequence of events involving AI agents highlights some key points:
- The autonomy of these systems may allow them to perform actions directly on services available on the internet;
- The identification of an agent's behavior does not necessarily mean that its origin can be determined securely;
- Attempts to access and effectively successful invasions are distinct situations and require different assessments;
- The actions of AI agents already involve public systems of more than one country, according to the events described in the source.
The Canadian investigation continues without a confirmed attribution of the attempts, while OpenAI examines the findings presented by the research company.
Valdir Antonelli is a journalist specializing in digital marketing and consumer behavior.
