“Anthropic's AI models could reach AI Safety Level 3 (ASL-3) in 2025.”