In an old fable from the Arabian Nights, a poor fisherman frees a genie from captivity only to discover the creature may turn on its liberator. Dario Amodei, chief executive of Anthropic, appears determined not to risk a similar fate. Or so goes the modern fable. His company chose not to publicly release ‘Mythos’, the newest and most powerful model in the Claude family of artificial intelligence (AI) systems, because it was considered too potent for open deployment. Anthropic itself described the model’s capabilities as “substantially beyond those of any model we have previously trained”.
