Anthropic stated in a November 2026 S-1/A filing update that internal estimates indicate Anthropic may imminently meet its recursive self-improvement danger threshold