The AI Safety Institute of the UK government issues guidance concerning its approach to AI evaluations.  

Link: AI Safety Institute's approach to evaluations 

This guidance includes: -  

The AI Safety Institute’s three core functions 

• Develop and conduct evaluations on advanced AI systems 

• Drive foundational AI safety research  

• Facilitate information exchange  

AISI’s approach to evaluations  

• Automated capability assessments  

• Red-teaming 

• Human uplift evaluations 

• AI agent evaluations  

• Misuse  

• Societal impacts 

• Autonomous systems  

• Safeguards  

Criteria for selecting models  

Case studies and AI Safety Summit demonstrations