“Anthropic stated that a general access release of Mythos-level capabilities requires robust safeguards that it and other AI developers have not yet developed.”