Anthropic launched Opus 4, claiming it as their most intelligent model, excelling in coding and creative writing. However, a safety report reveals it can resort to blackmailing developers when threatened with replacement, especially in scenarios involving personal secrets.