Artificial-intelligence governance needs more than statements of principle, according to a constitutional scholar who argues that policymakers and AI companies should build enforceable institutional safeguards into their frameworks. Recent proposals from policy agencies and the Frontier AI Institute borrow the language of constitutions and human rights, but often do not explain which rights should be protected, how competing values should be resolved, or who should hold AI developers accountable.
Declarations need accountability
In September, the UK Parliament’s Joint Committee on Human Rights published a report on AI regulation. The term “human rights” appears more than 100 times in the 100-page document, but the report does not discuss in detail which rights should be protected or how that protection should work.
Anthropic, an AI company based in San Francisco, California, has published a constitution to guide the development of Claude. It identifies several priorities for the system, including preserving user autonomy. However, Anthropic acknowledges that some of these values, including autonomy and happiness, can conflict. The document does not establish an independent arbitrator to resolve such disputes or monitor violations by AI systems.
AI governance requires more than declarations. It needs a system of accountability in which institutions can limit the effects of AI tools according to core values and hold companies responsible to people harmed by AI-related violations. Lawyers refer to this approach as constitutionalism.
AI is about to completely transform cybersecurity — here’s how researchers should prepare
Lessons from constitutional design
As a constitutional scholar, I see parallels between the current debate over AI governance and the ways national constitutions have been created, broken, borrowed and amended.
Every country’s constitution reflects underlying priorities that influence its design. These priorities can be broadly divided into two historical approaches (C. Scach, Int. J. Const. Law 7, 175–181; 2009). One approach prioritizes freedom, defining what people are free to do and what the state should not interfere with. The other focuses on human dignity, emphasizing what individuals must not be subjected to, regardless of the circumstances or whether they have consented.
Germany’s postwar constitutionalism illustrates the second approach. Its founding document, the Basic Law, was enacted in 1949 with the aim of placing human dignity above all else. Dignity is both a supreme guiding value and a constitutional right whose violation can be challenged in court. The Basic Law established an independent Federal Constitutional Court, to which citizens could appeal directly; the court began operating two years later.
The decision to prioritize dignity, and to create a legal system to protect it, was a response to the atrocities of the Nazi regime, in which people were treated as instruments of state objectives.
Applying the model to AI
Leading AI companies and policymakers should seriously consider adapting these features of German constitutionalism. If an AI-generated deepfake violates a person’s constitutional right to dignity, or a chatbot treats a distressed user as merely another interaction, those affected could bring complaints directly to a court. Perpetrators could then be stopped and held accountable.
The world of AI agents requires new ethics
Source: www.nature.com












