Corporations Constituting Intelligence — California Law Review

https://www.californialawreview.org/online/fragmented-democracy-bcpd6

Publish Date: 2026-03-30 12:00:00

Source Domain: www.californialawreview.org

Here is a summary of the article highlighting between 4 to 8 key points:

Unprecedented AI Constitution: Anthropic published a 79-page “constitution” for its AI model Claude, detailing the company’s intentions for the AI’s values and behavior in great depth, unlike prior lists of principles.
Philosophical Framework: The document acknowledges the uncertain moral status of AI models like Claude and establishes a hierarchy where Anthropic’s instructions take precedence, providing unique insights into corporate philosophy regarding artificial intelligence.
Conscientious Objection: The constitution permits and even encourages the AI to refuse certain requests when its value systems are threatened, drawing parallels to human professions with moral commitments.
Gaps in Ethical Application: The constitution does not apply to specialized AI models deployed by government or military, potentially conflicting with public ethical standards.
Democratic Deficit: Despite demonstrating a public participatory process that resulted in a constitution with significant divergence from its own, Anthropic’s final published constitution disregards public input.
Resistance and Compliance Issues: Anthropic resisted government restrictions but ultimately failed to prevent the military from embedding its AI model in critical operations, highlighting broader issues of regulation.
Governance Challenges: The constitution illustrates the need for public institutions to govern AI principles instead of relying on private entities to ensure ethical standards and democratic values.
Legitimacy vs. Thoughtfulness: While the document reflects thoughtful construction, it lacks democratic legitimacy, emphasizing the need for public mechanisms to authorize the principles governing AI behavior.