Altman releases OpenAI's five principles, hinting that future safety restrictions may limit the model's capabilities

According to Beating Monitoring, OpenAI CEO Sam Altman personally published five operational principles for the company, namely democratization, empowerment, universal prosperity, resilience, and adaptability. Altman previewed in “adaptability” that in the future, OpenAI might tighten user access to models for safety reasons during certain periods, explicitly stating that it can be imagined as “sacrificing some empowerment to gain greater resilience.”

“Democratization” requires that key AI decisions be made through democratic procedures and should not be solely decided by AI laboratories. “Empowerment” emphasizes giving users as much freedom to use as possible, but leans towards caution when uncertain. “Universal prosperity” defends OpenAI’s recent large-scale purchase of computing power and global data center construction, with Altman stating the goal of reducing AI costs to make it affordable for everyone, and that governments may need to explore new economic models to ensure value sharing. “Resilience” mentions a specific risk scenario: extremely powerful models could lower the barrier to creating new pathogens, requiring society-wide responses.

Altman reflected on the hesitation at the time to release the GPT-2 weights publicly, saying that in hindsight it was overthought, but it led to OpenAI’s current iterative deployment strategy.