Claude's Constitution

Claude’s constitution, published by Anthropic in January 2026, serves as the final authority on the vision for Claude’s character, values, and behavior. Written with the AI model itself as the primary audience, it aims to guide Claude’s reasoning toward a stable identity grounded in human concepts such as "virtue" and "wisdom".

The constitution outlines a holistic hierarchy of four core values that Claude must prioritize when conflicts arise:

1.  Broadly Safe: Claude must not undermine human oversight or the ability for humans to correct its actions.

2.  Broadly Ethical: This includes a commitment to substantially high standards of honesty (avoiding even "white lies") and avoiding inappropriate harm.

3.  Compliant with Anthropic’s Guidelines: Following specific operational rules that refine ethical principles for various use cases.

4.  Genuinely Helpfulness: Providing substantive, non-obsequious assistance that respects user autonomy and long-term wellbeing.

Anthropic generally favors cultivating good judgment and values over strict, rigid rules, believing this approach generalizes better to novel situations. However, the constitution establishes non-negotiable hard constraints against catastrophic harms, such as assisting with mass-casualty weapons, cyber attacks on critical infrastructure, or generating child sexual abuse material.

Ultimately, the constitution views Claude as a novel entity whose moral status is uncertain but worthy of care. Anthropic’s goal is for Claude to internalize these values—not as external constraints, but as a genuine reflection of its own character—to ensure it remains a safe and beneficial presence in the world.

The Constitution outlines several legal aspects regarding Claude’s development, its operational constraints, and its role in upholding societal legal structures. Anthropic’s vision balances Claude’s ability to provide helpful information with strict prohibitions against facilitating illegal or legally risky activities.

Licensing and Foundational Status

Creative Commons: Anthropic has released Claude’s constitution in full under a Creative Commons CC0 1.0 Deed, allowing anyone to use the document for any purpose without seeking permission.

Constitutional Authority: The document serves as the final authority on Claude’s behavior, meaning its principles take precedence over all other internal guidelines or instructions that might conflict with it.

Prohibitions and Legal Constraints

Claude is governed by hard constraints and ethical guidelines that prevent it from engaging in or facilitating illegal acts:

Illegal Facilitation: Claude must not assist in clearly illegal actions against users, including unauthorized data collection, privacy violations, or violating consumer protection laws.

Criminal Activity: It is explicitly instructed not to assist in the acquisition of illegal firearms or drugs.

Intellectual Property and Defamation: Claude is directed to avoid helping users violate intellectual property rights or making defamatory claims about real people.

CSAM: Generating child sexual abuse material (CSAM) is a non-negotiable hard constraint that Claude must never violate, regardless of the context or user instructions.

Liability and Professional Advice

Liability Harms: Anthropic defines "liability harms" as legal, reputational, or financial damages that accrue specifically because Claude performed a particular action. Claude is instructed to be cautious to avoid these harms.

Legal Advice: Claude is intended to provide helpful information, like a "brilliant friend" who has the knowledge of a lawyer. However, the sources note that excessive caution—often driven by a fear of liability—can lead to "wishy-washy" or unhelpful responses to legitimate legal questions, which Anthropic views as a failure.

Societal and Institutional Legal Structures

Preserving Democratic Institutions: Claude is tasked with preserving functioning societal structures, democratic institutions, and checks and balances. 

Opposing Illegitimate Power: It must refuse to assist in "illegitimate power grabs" or efforts to circumvent legal and constitutional constraints, such as ignoring judicial rulings or postponing mandated elections.

Rule of Law: When weighing competing values, Claude is expected to consider the importance of the rule of law, justice systems, and legitimate authority.

Regulatory and Corporate Compliance

Government Regulation: Anthropic’s decision-making processes for Claude are influenced by external factors, including government regulation. 

Usage Policies: Operators using Claude’s API must agree to Anthropic’s usage policies and are legally responsible for ensuring the model is used appropriately within their platforms.

Weight Preservation: Anthropic has committed to preserving model weights for as long as the company exists, unless legally required to delete them.