OpenAI announced a $1 million grants program to fund experiments in "democratic inputs to AI," seeking to shape how its models should behave within societal bounds [source]. The initiative follows internal acknowledgment that decisions about AI behaviour—what to refuse, how to respond to contested topics—cannot be made by the company alone.

The program invites proposals for mechanisms that could solicit and aggregate public input on AI rules and boundaries. OpenAI stated it aims to award ten grants of up to $100,000 each, with a focus on proof-of-concept designs rather than fully deployed systems. The company cited its existing usage policies as examples of decisions requiring broader input, noting that current approaches rely on internal teams and user feedback loops that may not represent diverse global perspectives.

OpenAI framed the challenge as analogous to content moderation at scale: whose values should govern when a model declines a request or frames a sensitive answer? The announcement referenced ongoing criticism that AI systems reflect narrow viewpoints, and positioned democratic input as a potential corrective.

The grants program does not commit OpenAI to implementing any resulting proposals. The company described the initiative as exploratory, intended to surface ideas rather than establish binding governance structures. No timeline was provided for evaluating submissions or integrating findings into model behaviour.

The announcement did not address how democratic processes would interact with OpenAI's existing content policy framework, nor whether such mechanisms would apply to all deployment contexts including enterprise and API customers. The program represents an acknowledgment that current methods for setting model boundaries lack transparent external input, though the scope and enforceability of any future democratic process remain undefined.