Corporations Constituting Intelligence

In January 2026, Anthropic published something unprecedented: a 79-page “constitution” for its AI model Claude. The document is remarkable. It is not a terms of service agreement. It is not a list of prohibited outputs. It is, as the company describes it, “a detailed description of Anthropic’s intentions for Claude’s values and behavior.” This constitution stands as a foundational text meant to shape how an artificial intelligence understands itself, its obligations, and its place in the world. Anthropic is not merely programming behaviors. It is, at least rhetorically, cultivating something like AI integrity. Reading the document, one encounters genuine philosophical engagement rather than boilerplate risk mitigation. History offers little reason to believe that corporate ethics survive contact with quarterly earnings reports. But a possible future in which Anthropic is incentivized to abandon its morals is not the only risk. My claim, and the focus of this piece, is stronger: Anthropic’s constitution itself harbors existing risks, found in both express limitations and structural defects.