“Select government entities and major banks were given early access to Anthropic's new AI model, Mythos, to test their own cybersecurity vulnerabilities.”