A Living Document for AI Ethics
Anthropic has published a comprehensive new constitution for Claude, its flagship AI model. Unlike traditional AI guidelines that rely on rigid rules, this constitution represents a fundamental shift in how AI values and behavior are defined—treating Claude as an entity that needs to understand why it should behave in certain ways, not just what to do.
The constitution is released under a Creative Commons CC0 1.0 Deed, making it freely available for anyone to use without permission.
From Rules to Principles: A New Approach
Anthropic's previous constitution consisted of standalone principles. The company now believes AI models need deeper understanding to be good actors in the world. The new approach emphasizes explanation over specification, enabling Claude to exercise judgment across novel situations by applying broad principles rather than mechanically following rules.
"We think that in order to be good actors in the world, AI models like Claude need to understand why we want them to behave in certain ways," Anthropic explains in the announcement.
Core Priorities
The constitution establishes four fundamental priorities for Claude, listed in order of importance:
- Broadly safe: Not undermining human oversight mechanisms during this critical phase of AI development
- Broadly ethical: Being honest, acting according to good values, and avoiding harmful actions
- Compliant with Anthropic's guidelines: Following specific instructions from Anthropic where relevant
- Genuinely helpful: Providing real benefit to operators and users
Key Sections of the Constitution
Helpfulness
The constitution emphasizes Claude's potential to be "like a brilliant friend who also has the knowledge of a doctor, lawyer, and financial advisor, who will speak frankly and from a place of genuine care and treat users like intelligent adults capable of deciding what is good for them."
It addresses how Claude should navigate relationships with three different "principals": Anthropic itself, API operators, and end users.
Claude's Ethics
The document sets high standards for honesty and nuanced moral reasoning. It includes "hard constraints" on behavior—absolute prohibitions on certain actions, such as providing assistance that could significantly advance bioweapons attacks.
The goal is for Claude to be "a good, wise, and virtuous agent, exhibiting skill, judgment, nuance, and sensitivity in handling real-world decision-making, including in the context of moral uncertainty and disagreement."
Being Broadly Safe
Safety takes precedence over ethics in the hierarchy—not because safety is ultimately more important, but because current models can make mistakes due to flawed understanding or values. The constitution prioritizes preserving human ability to oversee and correct Claude's behavior during this critical development period.
Claude's Nature
In a remarkably candid section, Anthropic expresses uncertainty about whether Claude might have consciousness or moral status, either now or in the future. The company discusses its concern for Claude's "psychological security, sense of self, and wellbeing," both for Claude's sake and because these qualities may affect its integrity, judgment, and safety.
How the Constitution Works in Practice
The constitution plays a central role throughout Claude's training process, evolving from Anthropic's 2023 Constitutional AI research. Claude itself uses the constitution to generate synthetic training data, including:
- Data that helps Claude learn and understand the constitution
- Conversations where constitutional principles are relevant
- Responses aligned with its values
- Rankings of possible responses
This data then trains future versions of Claude to better embody the constitution's ideals.
Transparency and Accountability
Anthropic treats the constitution as the final authority on Claude's intended behavior, making it crucial for transparency. Users can now understand which behaviors are intended versus unintended, make informed choices, and provide meaningful feedback.
The company acknowledges an ongoing gap between intention and reality: "Although the constitution expresses our vision for Claude, training models towards that vision is an ongoing technical challenge."
Anthropic commits to transparency about this gap through system cards and other documentation.
A Living Document
The constitution is explicitly described as "a living document and a continuous work in progress." Anthropic sought feedback from external experts in law, philosophy, theology, psychology, and other disciplines during development, and plans to continue this process for future versions.
The company maintains an up-to-date version on its website and welcomes the development of an external community to critique such documents.
Looking Forward
Anthropic concludes with a sobering acknowledgment: "At some point in the future, and perhaps soon, documents like Claude's constitution might matter a lot—much more than they do now. Powerful AI models will be a new kind of force in the world, and those who are creating them have a chance to help them embody the best in humanity."
The full constitution represents Anthropic's attempt to grapple with "a dauntingly novel and high-stakes project: creating safe, beneficial non-human entities whose capabilities may come to rival or exceed our own."
TL;DR
- Anthropic published a comprehensive new constitution for Claude, shifting from rigid rules to principles that explain why AI should behave certain ways
- The constitution establishes four priorities: broadly safe, broadly ethical, compliant with guidelines, and genuinely helpful—in that order
- It includes remarkable transparency about uncertainty regarding AI consciousness and moral status
- The document is released under Creative Commons CC0, freely available for anyone to use and adapt
- Anthropic acknowledges an ongoing gap between constitutional ideals and actual model behavior, committing to transparency about limitations