The Department for Science, Innovation and Technology share a research and analysis paper on the capabilities and risks of frontier AI. This paper outlines the capabilities and challenges posed by technology – such as misuse, social harms and loss of control.  

Link: Frontier AI  

This resource includes: -  

• Capabilities and ricks from frontier AI: discussion paper (PDF and HTML)  

• Future risks of frontier AI (Annex A) (PDF and HTML) 

• Safety and security risks of generative artificial intelligence to 2025 (Annex B) (PDF and HTML) 

Frontier AI Discussion Paper 

Introduction 

What is the current state of frontier AI capabilities? 

• How frontier AI works 

• Frontier AI can perform many economically useful tasks  

• Frontier AI could be more capable than evaluations indicate  

• Limitations of Frontier AI  

How might frontier AI capabilities improve in the future? 

• Recent AI progress has been rapid  

• Recent progress was driven by systematic trends in compute, data and algorithms 

• Scaling laws: performance improves predictably with increased compute and data 

• Rapid AI progress is likely to continue for several years  

• Advanced general-purpose AI agents might be developed in the future  

What risks do frontier AI present?  

• Cross-cutting factors  

• Is it difficult to design sage frontier models in open-ended domains 

• Evaluating the safety of frontier AI systems is an open challenge  

• It may be difficult to track how frontier AI systems are deployed or used  

• AI safety standards have not been established  

• Insufficient incentives for AI developers to invest risk into mitigation measure  

• There may be a significant concentration of market power in AI  

Societal Harms 

• Degradation of the information environment  

• Labour market disruption 

• Bias, Fairness and representational harms 

Misuse risk 

• Dual Use Science risks 

• Cyber 

• Disinformation and influence operations 

Loss of control 

• Humans might increasingly hand over control to misaligned AI Systems 

• Future AI systems might actively reduce human control 

Future Risks of Frontier AI (Annex A) 

• Current Frontier AI Capabilities  

• Future Frontier AI Capabilities  

• Other Critical Uncertainties  

• Future Scenarios for Exploring Frontier AI Risks  

• Frontier AI Risks  

Safety and security risks of generative artificial intelligence to 2025 (Annex B) 

• The generative AI ecosystem  

• Threat actors 

• Safety and security risks