At the Ai4 conference in Las Vegas, three prominent AI researchers—Geoffrey Hinton, Fei-Fei Li, and Andrew Ng—argued that growing concerns about AI safety should not automatically lead to highly restrictive access to AI technology. Although the three differ on specific regulatory approaches, they share a concern that excessive restrictions could concentrate AI development in the hands of a small number of powerful companies. Andrew Ng, in particular, warned against creating “gatekeepers” that determine who can access and build with AI.
Their argument centers on the value of openness for innovation and competition. If only a handful of major AI companies control advanced models and infrastructure, those companies could effectively determine the direction of AI development. Open models, research and broader access can instead allow universities, startups and independent developers to experiment with the technology, challenge dominant players and create applications that large companies may not pursue. This is especially relevant as AI becomes an increasingly important general-purpose technology.
However, the debate is complicated by genuine AI safety concerns. More capable models can introduce risks involving misuse, cybersecurity, misinformation and autonomous behavior. The challenge is determining which capabilities should remain broadly available and which may require additional safeguards. Hinton, Li and Ng do not simply dismiss these risks; rather, their position is that safety measures should be designed carefully enough that they do not unintentionally give a small group of companies permanent control over AI development.
The broader issue is therefore a difficult balance between AI safety and AI openness. Keeping AI accessible can encourage competition, research and distributed innovation, while unrestricted access to increasingly capable systems could create new risks. The three researchers' position suggests that the solution may not be to choose one side completely, but to develop safeguards that target genuinely dangerous capabilities while preserving broad access to useful AI. As the technology becomes more powerful, the question may increasingly be not simply how safe should AI be?, but who gets to decide what “safe” means and who is allowed to build with AI.
