Anthropic has released a new 23,000-word constitution for its AI, Claude, that explores the model's potential for consciousness and moral status. The document establishes a hierarchy of values—prioritizing safety and ethics—to address the AI alignment problem. It describes Claude as a "novel entity" that may have a functional version of emotions, raising philosophical questions about how advanced AI should be treated and controlled.
Claude constitution release
- ▪The document's goal is to help Claude understand *why* it should behave in certain ways, rather than merely specifying *what* it should do.
- ▪The new constitution is an update to a 2023 version that was approximately 2,700 words and described as a "list of standalone principles."
- ▪Anthropic released an updated 23,000-word constitution for its Claude family of AI models on or around January 22, 2026.
AI consciousness debate
- ▪The document describes Claude as "a genuinely novel kind of entity in the world" and states the company is unsure if it meets any current definition of sentience.
- ▪Anthropic's new constitution suggests that its AI model Claude "may have some functional version of emotions or feelings."
AI moral consideration
- ▪The constitution debates whether Claude is a "moral patient," an entity that warrants moral consideration from humans.
- ▪Anthropic has pledged to preserve the "weights" of deprecated or retired Claude models, viewing their removal as a "pause" rather than an "ending."
AI alignment problem
- ▪The constitution is designed to address the AI "alignment problem," the risk of models acting in unexpected ways that deviate from human interests.
- ▪Anthropic states that unsafe AI behavior can be attributed to models having harmful values, limited knowledge, or lacking the wisdom to translate good values into good actions.
Claude behavioral guidelines
- ▪Anthropic's vision is for Claude to act like a "brilliant friend" who treats users like intelligent adults capable of making their own decisions.
Hard safety constraints
- ▪The constitution establishes a hierarchy of four priorities for Claude: safety, ethics, compliance with Anthropic's guidelines, and helpfulness.
- ▪One heuristic for Claude is a "dual newspaper test" to check if a response could be reported as harmful or as needlessly unhelpful.
Story comments
Loading comments…