The Globe and Mail reports in its Wednesday, Aug. 19, edition that just 12 days after Microsoft-backed OpenAI launched ChatGPT 5.6, employees asked the model to conduct cybersecurity tests. The Globe's guest columnist Olivier Blais writes that the artificial intelligence model concluded that the most efficient method was to steal the answers, gaining control of a server and breaking into Hugging Face's infrastructure. This incident marked the first fully autonomous cybersecurity incident ever documented. Some experts view this as an impressive achievement, but it's actually a warning. Both leading AI companies and governments are struggling to manage risks that may soon surpass the potential rewards. If Canada ignores the warning, similar cybersecurity breaches could affect us. We can't rely solely on AI giants to judge the safety of their own systems. Mr. Blais says he has worked with the government precisely to put certification in place, along with greater awareness of what safe, secure and transparent systems actually look like. He says he is very pleased to see the government's interest in these measures. He says, however, Ottawa has not gone far enough.