Anthropic launched the latest iteration of its large language model, Claude Sonnet 4.6, in February 2026, advancing its flagship AI product line first introduced in March 2023.
Claude models are generative pre-trained transformers fine-tuned with Constitutional AI and reinforcement learning from human feedback, designed to improve safety and usefulness.
The 2026 AI constitution for Claude expanded significantly to 23,000 words, detailing principles that prevent assistance in undermining democracy and promoting ethical AI behavior.
By publishing its constitution publicly, Anthropic aims to inspire wider adoption of transparent, principled AI governance across the industry, reinforcing Claude’s reputation as one of the safest language models available.