As AI capabilities develop, so too are the measures to ensure safety and control, according to the inaugural trends report from the UK government-backed AI Security Institute.

The Frontier AI Trends Report is a public assessment of how the most advanced AI systems are evolving and draws on two years of testing AI capabilities in areas critical to innovation and security, including cyber, chemistry and biology.

It found that safeguards to ensure AI behaves as intended are improving, with work continuing between government and companies to strengthen them and ensure the full potential of AI can be unlocked while keeping people safe. While every system tested remains vulnerable to some form of bypass and protection still varies across companies, it says huge strides are being made. The time it took AISI red-teamers to find a ‘universal jailbreak’ – a general way of getting round a model’s safety rules – for example, increased from minutes to several hours between model generations.

AISA says he report is not intended to make policy recommendations, but gives decision makers in the UK and internationally clear data on what the most advanced AI systems can do. Its key findings suggest that the most advanced AI systems are improving at remarkable speed, having gone from struggling with basic tasks to matching or surpassing human experts in some areas in two years.

In the field of cyber security, it found success on apprentice-level tasks has risen from under 9% in 2023 to around 50% in 2025. For the first time in 2025, a model completed an expert level cyber task, requiring up to 10 years of experience. In biology and chemistry, the report suggests that systems are now outperforming PhD-level researchers on scientific knowledge tests and helping non-experts succeed at lab work that would previously have been out of reach.

AI minister, Kanishka Narayan, said: “This report shows how seriously the UK takes the responsible development of AI. That means making sure protections are robust, and working directly with developers to test leading systems, find vulnerabilities and fix them before they are widely used.

“Through the AI Security Institute, we are building scientific capability inside government to understand these systems as they evolve, not after the fact, and to raise standards across the sector. This report puts evidence, not speculation, at the heart of how we think about AI, so we can unlock its benefits for growth, better public services and national renewal while keeping trust and safety front and centre.”

---