Anthropic’s new Claude Mythos AI model has apparently found thousands of vulnerabilities in ‘every major operating system and every major web browser, along with a range of other important pieces of software’