Anthropic Tests Claude Mythos With Early Access

Anthropic is testing Claude Mythos, its reportedly most powerful AI model, with a select group of early-access customers, the company and a data leak said on April 2, 2026. Anthropic says Mythos improves reasoning, coding and cybersecurity capabilities, but a publicly available draft warned of "unprecedented cybersecurity risks" and experts raised concerns about jailbroken models and oversight.