Frontier AI Matches Human Cybersecurity Pros: UK Report

Frontier artificial intelligence models are now matching human cybersecurity professionals in capabilities. This is according to new benchmark data from the UK AI Security Institute. What's interesting is that these AI models, like Anthropic Claude Sonnet 4.5, can complete tasks in a fraction of the time human experts take. For example, a task that typically requires 16 minutes of human labor can be done by AI with 80 percent reliability. The Institute's data, published in May 2026, shows the capability doubling period for these AI systems has fallen to about four months. This acceleration could help fill millions of unfilled cybersecurity roles globally. However, it also threatens to displace entry-level and mid-tier analysts. This means human cybersecurity professionals will likely need to shift focus to high-level strategic work. This rapid AI advancement demands a massive global reskilling initiative for the workforce.