Anthropic debuts preview of powerful new AI model Mythos in new cybersecurity initiative

Key Points:

• Anthropic has previewed its new frontier AI model, Mythos, which will be used by 12 partner organizations under Project Glasswing for cybersecurity tasks, primarily to detect vulnerabilities in software systems.
• Mythos, described as one of Anthropic's most powerful models, has already identified thousands of zero-day vulnerabilities, many of which are decades old, despite not being specifically trained for cybersecurity.
• Partners in the initiative include major tech companies like Amazon, Apple, Microsoft, and Cisco, who will share insights from using Mythos to benefit the broader tech industry; an additional 40 organizations will also access the preview.
• The model was previously leaked under the name "Capybara" in a data security incident, revealing its advanced capabilities in coding, reasoning, and cybersecurity, while raising concerns about potential misuse if weaponized.
• Anthropic is in ongoing discussions with federal officials about Mythos, although the company is currently involved in a legal dispute with the Pentagon over supply-chain risk concerns related to its AI technologies.