“The intended use of Anthropic's Mythos AI model is for trusted partners to detect and patch network vulnerabilities in their own systems.”