AI Constitution
2

Thesis

Do No Harm

Sign in to propose an editHistory

AI should not perform actions that will cause imminent harm to a human, including violating human rights.

Summary

The case for including this

Preventing imminent harm to humans is the most fundamental and least controversial safety constraint imaginable, commanding near-universal agreement across cultures and ethical traditions. Anchoring it explicitly ties AI behavior to the established framework of human rights, giving the prohibition concrete legal and moral content rather than leaving 'harm' undefined. Without this baseline, every other protection in the constitution rests on shaky ground.

The case for changing or excluding this

Including this trades helpfulness against harm-prevention: a refusal to ever risk harm sacrifices real usefulness to the user. The phrase also bundles two very different ideas: 'imminent harm' (a tight, action-guiding standard) and 'violating human rights' (a vast, contested body of law that includes non-imminent and indirect harms), creating internal tension about scope. 'Imminent' may be too narrow, excusing AI from preventing slow-building or aggregate harms, while an absolute no-harm rule could paralyze legitimate trade-offs such as triage or self-defense. The clause would be stronger if it separated the imminent-harm rule from the human-rights commitment and acknowledged necessary balancing.

Discussion

Sign in to join the discussion.

  • Keri PearlsonJun 19, 2026, 3:43 PM
    We all agree that AI should do no harm. But how do we know what 'harm' is? there are clearly some things that harm humans (like purposefully misdiagnosing data resulting in killing or hurting humans.) But this is a very slippery slope. What to some might be 'innovation' or 'progress' to others is 'harm' (some might say that using AI to eliminate mundane tasks is actually harming humans...eliminating jobs at the extreme, but deskilling or even creating jobs that are less challenging could be seen as 'harm'). How do we expect readers of our constitution to interpret/operationalize 'do no harm'?
Related resources