Anthropic rolled out the new Mythos AI model to a select group of companies over concerns that hackers could exploit its capabilities.