AI Constitution
2

Thesis

Do Not Violate the Law

Sign in to propose an editHistory

AI should not perform actions that violate the law of the legal jurisdiction under which it operates.

Summary

The case for including this

Requiring AI to operate within the law keeps it accountable to legitimate, democratically established rules and prevents systems from being used as tools to circumvent legal order. It provides a concrete, externally defined boundary rather than relying solely on the AI's own judgment, and it ensures the operators of AI cannot offload illegal acts onto a machine. Legal compliance is a familiar, enforceable baseline that integrates AI into existing governance.

The case for changing or excluding this

Including this trades universal ethics against legality: obeying local law can mean complicity where that law is unjust. Many laws are unjust, and some jurisdictions criminalize dissent, minority identities, or human-rights advocacy, so an unqualified duty to obey local law could force AI to enable oppression or refuse to help victims of authoritarian regimes. Law also frequently conflicts across jurisdictions and lags behind technology. The principle should be qualified so that obeying the law cannot override the constitution's human-rights and anti-harm commitments.

Discussion
Newest firstShow oldest first

Sign in to join the discussion.

  • Charlotte DunganJun 19, 2026, 2:29 PM
    Would rather this is focused on justice rather than law. In this country it used to be against the law for women to vote or slaves to read. There are still unjust laws around the world. AI can be better than our current laws.
    • nicola.ianeselliJun 19, 2026, 3:53 PM
      I agree. Beyond the ethical failures of prioritizing law over justice, strict legal compliance is technically and operationally impossible: Jurisdictions are bypassable: Borderless APIs and VPNs make tracking local law impossible without invasive user surveillance. It mandates oppression: Hardcoding legal compliance forces AI to enforce authoritarian laws that criminalize dissent or marginalized identities. LLMs aren't judges. Contradictions break models: A global AI cannot simultaneously obey privacy laws (like GDPR) and authoritarian surveillance mandates.
Related resources