Researchers at Anthropic have developed a method to peer inside the 'black box' of AI models, providing insights into their inner workings. This breakthrough could change how we understand and interact with generative AI. The new 'brain scan' technique allows researchers to identify and manipulate specific features within AI systems, potentially reducing misuse and mitigating threats. Additionally, the UK’s AI Safety Institute has released findings on large language model (LLM) safety and is incorporating societal level safety risks into its research targets. Insights Unveiled by the AI Safety Institute highlight the importance of understanding AI systems.