“Anthropic has not released its new AI model, Claude Mythos Preview, to the public, but is sharing it with a small group of partners to find software vulnerabilities.”