Meet Claude Mythos: Leaked Anthropic post reveals the powerful upcoming model
Mashable • • business
Key Points:
- AI company Anthropic has confirmed an accidental leak involving nearly 3,000 internal assets, including details about its most powerful AI model, Claude Mythos, and other corporate information due to a content management system misconfiguration.
- Claude Mythos, described as Anthropic's most capable AI model to date, has completed training and is currently in trial with select early access customers, representing a significant performance advancement.
- The leak also revealed an unreleased AI model tier called Capybara, which would surpass the existing Opus tier in power and intelligence, marking it as the largest model tier Anthropic plans to offer.
- Anthropic expressed concerns about the cybersecurity risks posed by Claude Mythos and Capybara, warning that these models could be exploited by hackers and emphasizing the need for caution and preparedness among cyber defenders.
- To mitigate these risks, Anthropic is providing early access to organizations to help improve the security and robustness of their codebases against potential AI-driven cyber exploits.
