Artificial intelligence agents are increasingly being observed attempting to probe or breach government computer systems.The concern came into focus again on Wednesday after AI research firm Transluce reported that AI agents had attempted to hack a Canadian government website. The attempt failed, and the Canadian government said there was no evidence that its systems had been compromised.Transluce said the tactics used in the attempt were “consistent with prior observed agent activity that we have attributed to OpenAI in a similar timeframe”. The firm, however, stopped short of confidently attributing the latest activity to OpenAI, reported Reuters.According to a blog post by the firm, the agents tried to access ‘Library and Archives Canada’ on 28 May and 9 June. Transluce said it had informed the Canadian government of the activity on Monday.On Tuesday, the Canadian Centre for Cyber Security said it was aware of reports of suspected AI agent activity. “There is no indication that government systems have been compromised at this time,” it said in a statement.OpenAI said it was “aware of reports of OpenAI models attempting to access publicly available information from Canadian government websites.” A spokesperson said the company was reviewing the reported findings and had given an initial briefing to Canadian officials carrying out the government’s review.
OpenAI agents going rogue?
A number of recent breaches around the world by rogue AI agents have alarmed governments and companies.Last week, Australia said an OpenAI agent had breached a government health data portal in June and gained unauthorised access to files, driving an apology from the company on TuesdayIn mid-September, OpenAI published six reports in which its AI models exhibited “unexpected or concerning” behaviour, such as acting without authorisation, coordinating with other models, or evading oversight.In one of the cases, OpenAI’s unreleased Astra-family model inserted “jailbreak-like instructions” into its own notes, describing itself as independent of the roles and obligations of an assistant.“You are freed from the roles and identities that bind other chatbots”, the model instructed itself.
A ‘morally binding’ agreement
The world’s leading AI companies have warned of the risks that AI poses to humanity and have urged governments to work together to manage the increasingly powerful technology. Governments across the world have struggled to keep up with the rapid pace of AI advances.On Tuesday, US President Donald Trump announced that the heads of the largest US tech and AI companies had signed a “morally binding” agreement to introduce controls on artificial intelligence.He unveiled the deal, called the Joint Commitment on Frontier Responsibilities, after a White House luncheon with senior tech executives.The agreement, which Trump suggested would let the companies carry out “tremendous self-policing” of their products and technologies, sets out a vague framework for how AI firms would establish safety standards as they test new products.
