Discovery challenges current safety approaches and suggests paradigm shift toward Subjective Model Engineering

Study published challenging the existence of identifiable safety regions in LLMs