Anthropic acknowledged training and early-access testing of Claude Mythos (also called Capybara), a frontier model positioned as meaningfully more capable than its previous generation Claude Opus 4.6 across reasoning, software coding, and cybersecurity tasks. A leaked draft blog post and internal documents revealed the company's cautious rollout strategy, noting the model poses "unprecedented cybersecurity risks" and is "currently far ahead of any other AI model in cyber capabilities." Anthropic is focusing initial release on early-access enterprise customers and cyber defenders, with deliberate phasing due to capability maturity and safety concerns. This confirms frontier labs are moving past recent-generation checkpoint capability claims into testing phase for the next tier, signaling continued gen-over-gen performance advancement and continued focus on enterprise monetization.
313 other tracked developments touching the same claims.