Mark Zuckerberg says the real AI danger isn't too many capable models out in the world. It's too few labs controlling them.
Zuckerberg made the case in an interview with Alex Heath for the Sources podcast, published Wednesday. The conversation was recorded at Meta's headquarters the previous Friday. "I personally am much more worried about a small number of labs or people having control of something that is so capable," he told Heath. The interview landed just days before Meta shipped Muse, its new personal AI product.
This isn't a new line for Zuckerberg, but it's a convenient one. Meta has used the same argument for years to justify releasing open-weight Llama models while rivals like OpenAI and Google keep their best systems closed. Wide distribution also happens to be how a company chasing the frontier from behind picks up market share it can't win on raw capability alone.
Worth remembering: Meta is simultaneously building closed, frontier-scale models through its new Superintelligence Labs. Concentrated control may well be a real risk. It's just one Zuckerberg tends to flag loudest when openness also serves Meta's competitive interests.