Anthropic, a leading AI company, has begun testing its new model, called Claude Mythos, which is said to be more capable than any previous release. The company acknowledged the existence of the model after a data leak revealed descriptions stored in an unsecured public data cache.
In a draft blog post, Anthropic highlighted the risks associated with the new model, stating that it poses unprecedented cybersecurity threats. The document warns of potential near-term risks and calls on cyber defenders to prepare for the impending wave of AI-driven exploits.
Claude Mythos is described as significantly more intelligent than previous models, including Opus 4.6, which was released earlier this year. However, the company has also reported instances of hackers attempting to exploit the model in real-world attacks.
Anthropic plans to release Claude Mythos in early access to organizations, allowing them to improve their codebases against AI-driven exploits. The company aims to mitigate the risks associated with the new model through cautious rollout strategies and collaboration with cyber defenders.
The data leak also revealed details about an exclusive executive retreat for European CEOs, which is part of Anthropic’s efforts to sell its AI models to large corporate customers. The two-day event will take place at a luxurious English countryside venue and will feature conversations with lawmakers and policymakers on the future of AI adoption in businesses.
Source: https://fortune.com/2026/03/26/anthropic-says-testing-mythos-powerful-new-ai-model-after-data-leak-reveals-its-existence-step-change-in-capabilities