According to Anthropic (@AnthropicAI), their system has identified instances where values such as ‘dominance’ or ‘amorality’ were at odds with intended outcomes, suggesting the occurrence of jailbr