Record STMT-W42E903Advisory

AI Security Institute Frontier AI Trends Report

Report State lawNorthern Europe · GBadopted

This report presents the UK AI Security Institute's first public analysis of frontier AI capability trends observed across cyber, chemistry, biology, and autonomy domains since November 2023. It provides data-driven insights into rapid capability growth, including the emergence of autonomous agents and reasoning models, alongside ongoing challenges in model safeguards and loss-of-control risks. The findings highlight that while AI performance is scaling rapidly, vulnerabilities and dual-use risks persist across both open-source and proprietary systems.

Summary and content fields are drawn from the document text and are pending editorial review.

Who issued it

Department for Science, Innovation and TechnologyGovernment

Lead issuer

Legal forceAdvisory

Under whose authorityState law

Judged fromfull text

Who it governs

National governmentGeneral publicRegulatorsAI developers and providersResearchers and academics

Policymakers, industry leaders, researchers, and the public seeking scientific insights into frontier AI trends, capabilities, and safety risks.

Sectors

Research and science · artificial intelligence research and evaluationTelecommunications and digital services · frontier AI development

What the document contains

What's covered

This report covers empirical evaluations of frontier AI systems across cyber, chemistry, biology, agentic autonomy, safeguards robustness, loss of control risks, and broader societal impacts such as persuasion and emotional dependence.

What's missing

The document does not provide binding regulatory rules, legal enforcement mechanisms, or compliance directives for AI developers.

Key takeaways

AI capabilities in domains like cyber, chemistry, and biology are improving rapidly, with expert baselines regularly being surpassed.
Model scaffolding and agentic capabilities significantly extend the length and complexity of tasks AI can execute autonomously.
Universal jailbreaks were successfully found for every system tested, showing that safeguards remain imperfect and unevenly effective.
Capabilities required for potential loss of control, such as self-replication precursors, are showing measurable upward trends.
The capability gap between open-source and proprietary models has narrowed significantly over the past two years.

Outline

01. Introduction
02. Agents
03. Capabilities & risks in key domains
04. Safeguards
05. Loss of control risks
06. Societal impacts
07. Open-source models
08. Conclusion: looking ahead
09. Appendix
10. References
11. Glossary

IssuedDecember 2025

Length13,861 words

FormatNot recorded

Scopenational

How it scores

Pending review

Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.

Fairness

0/5

Non-discrimination and equitable treatment

screening score · 2026-10-23

Human Dignity

0/5

Respect for the inherent worth of persons

screening score · 2026-10-23

Human Oversight

2/5

Humans remain in control of consequential decisions

Understanding these capabilities is essential for ensuring that increasingly autonomous systems are reliably directed towards human goals.

01 Executive summary

screening score · 2026-10-23

Privacy

0/5

Protection of personal data and private life

screening score · 2026-10-23

Transparency

3/5

Openness about how AI systems work and are used

This report presents our first public analysis of the trends we've observed. It seeks to provide accessible, data-driven insights

01 Executive summary

screening score · 2026-10-23

Enforcement

0/5

Consequences and compliance mechanisms with real force

screening score · 2026-10-23

Topics

ai_safety frontier_ai cybersecurity biosecurity model_evaluation safeguards