AI only through laws, companies are beginning to build “constitutions” directly into AI systems. These are sets of rules or principles that guide how an AI behaves—what it should say, avoid, or prioritize. For example, companies like Anthropic are training models such as Claude using a written “constitution” so the AI can evaluate its own responses and align with values like being helpful, honest, and harmless.
The article argues that this approach reflects a deeper shift: governance is moving from public institutions to private tech firms. Traditionally, rules about behavior—especially those affecting society—are decided through democratic systems like constitutions and laws. But with AI, companies are effectively writing their own rulebooks and embedding them into systems that millions of people interact with daily. This raises concerns about accountability, since these “AI constitutions” are not created through public debate or oversight.
Another key point is that these AI constitutions are not neutral—they reflect the values and assumptions of their creators. Even if designed with good intentions, they can shape how information is presented, what viewpoints are emphasized, and how decisions are made. In that sense, they function almost like invisible governance systems, influencing users without the transparency or legitimacy of traditional political frameworks.
Ultimately, the article raises a fundamental question: if AI systems are becoming powerful actors in society, who gets to decide their rules? The idea of an AI “constitution” may help make systems safer and more predictable, but it also highlights a growing tension—whether the future of governance will remain in democratic institutions or increasingly shift into the hands of technology companies designing the systems we rely on.
