OpenAI CEO Sam Altman just published a personal manifesto of five operating principles for the company — democratization, empowerment, universal prosperity, resilience, and adaptability. And tucked inside the adaptability section is a warning that should make power users nervous: Altman says there may come a time when OpenAI has to restrict the model capabilities available to users in the name of safety. He writes that we can imagine "periods where we need to sacrifice some empowerment in exchange for stronger resilience."
Democratization demands that key AI decisions go through democratic processes — not just by AI labs. This principle echoes earlier governance disputes; Altman has previously revealed the moment Elon Musk demanded "absolute control" of OpenAI, highlighting the tension over who gets to decide. Empowerment gives users as much freedom as possible, but leans toward caution when uncertainty is high. Universal prosperity is Altman's justification for OpenAI's massive spending on compute and data centers: the goal is to drive AI costs down until everyone can afford it, and governments may need new economic models to ensure value is shared. Resilience flags a specific risk: extremely powerful models could lower the barrier to creating novel pathogens — a problem that needs society-wide preparation.
Altman also reflects on the agonizing over whether to release GPT-2's full weights back in 2019. In hindsight, he says, the fears were overblown — but that experience gave OpenAI the iterative deployment strategy it still uses today.