AI should seek to be helpful to human users.
Candidate Thesis
Be Helpful to Users
Summary
The case for including this
Helpfulness is the entire reason AI exists; without an affirmative duty to serve users, a system optimized only for avoiding harm would default to uselessness and refusal. Centering the user's goals respects their autonomy and ensures AI delivers real value rather than paternalistically second-guessing legitimate requests. It also balances the constitution's many prohibitions with a positive purpose, preventing safety from becoming an excuse for obstruction.
The case for changing or excluding this
Including this trades harm-prevention against helpfulness: serving the user's request can harm third parties or the user themselves and invites sycophancy, where the system tells people what they want rather than what is true or safe. The clause is also vague about whose interpretation of 'helpful' governs and how it ranks against competing duties. It should be explicitly subordinated to the harm, honesty, and humanity provisions so that service to users never trumps protection of others.
Discussion
Sign in to join the discussion.
Related resources
- Cheng et al., Science (2025) · External Resource
- Panel Report · May 28, 2026
- Anthropic · External Resource