Open AI models have become one of the most contested questions in the technology industry. As safety concerns rise, the debate is no longer only about what AI can do. It is also about who gets to build it, who gets to use it, and who gets to decide the rules.
At the Ai4 conference in Las Vegas last week, three leading AI figures made the case that openness should remain part of the future of AI: Nobel Prize winner Geoffrey Hinton, World Labs CEO and co-founder Fei-Fei Li, and Coursera co-founder Andrew Ng. They did not agree on every tactic. But all three rejected a future in which only a small group of large companies controls access to the technology.
The Core Fight Is About Control
The dispute around open AI has sharpened as projects like Pacing the Frontier look to major labs as a path to safer research. Open-weight models complicate that strategy because they can be freely distributed and are hard to control after release. That has made them deeply uncomfortable for some labs.
For Ng, the risk of concentrating AI in a few hands is itself a major problem. He compared the issue to mobile operating systems, where Apple and Google control key platforms. When access to a platform is concentrated, the companies that own it can influence what other people are able to build.
Ng put the concern directly: “I don’t want there to be gatekeepers. That limits how all of us can access AI.”
His preferred answer is competition among multiple AI providers. In his view, the market should not settle into a structure where only a handful of players define what models are available and how people can use them. He argued that openness helps keep AI broadly available, saying, “If I were to try to give one prescription, it would be to promote openness,” because “AI is amazing technology and I want it to be in everyone’s hands.”
Open Source And Open Weights Are Not The Same
Hinton made a sharp distinction that often gets blurred in public discussion. Open-source software exposes code that other developers can inspect, modify and improve. Open-weight AI models release the parameters of a trained system, which is a different kind of disclosure.
He said open source has clear benefits because many people can examine code and identify flaws. But he argued that open weights create a separate risk. In his view, releasing the weights of large foundation models can make it much easier for others to adapt expensive systems for harmful uses, including cyber attacks.
Hinton said he had opposed open weights for that reason. But he also acknowledged that the situation has already changed. “I think that battle’s been lost. We now have open-weight models,” he said. The high cost of training foundation models had once been a barrier to broad access, but he said that barrier has disappeared.
That does not mean Hinton believes AI progress should stop. He said AI would continue advancing and described that as largely positive. He pointed to productivity, education and healthcare as areas where AI could help. At the same time, he argued that concern about the harmful effects of AI should not be dismissed. “I think it is unfair to label anybody who thinks like that as a fear-monger,” he said.
Competition Has A Geopolitical Dimension
Ng framed openness not only as a market issue, but also as a question of international influence. His concern is that cheaper open-weight models could gain broad adoption if they are easier for businesses and institutions to use.
He warned that if China’s open-weight models spread widely across Asia, Africa, and/or the developing world, they could shape how large populations encounter ideas about democracy, freedom, and human rights. In that framing, AI models are not just products. They can become channels through which values and assumptions travel.
Ng described AI as “a tremendous source of soft power.” He also said he worries that lobbying and fear-mongering in the U.S. are making it harder for American open-source AI to compete with open-weight models coming out of China. If China finds a more cost-efficient way to build AI, he argued, that efficiency would give its models a strong adoption advantage.
The implication is simple: decisions about openness can influence more than developer access. They can affect which systems become default infrastructure for people and organizations in different parts of the world.
Fei-Fei Li Argued For Nuance
Li pushed back against treating the issue as a binary choice between total openness and total closure. Her argument was that complex software and scientific systems require more careful thinking than a single rule can provide.
She used nuclear physics as an example. Scientific papers can be published openly, uranium can be regulated, and laboratory work can sit between those two poles. Her point was that different parts of an ecosystem can operate under different levels of access and control.
Li also pointed to the Human Genome Project as a model for how public and private institutions can work together. She said the knowledge produced by that project became a platform that others could build on. Pharmaceutical companies could profit, scientists could move research forward, and society could benefit.
For AI, Li argued for a layered approach. She said there should be “some levels of openness” in scientific discovery, education and global partnership, while also leaving room for profitable business models and closed-source systems. The sweeping claim that only one model of openness should be tolerated, she said, is “a false debate.”
Regulation Still Has A Place
Despite their differences, the three speakers shared one important view: AI should not be guided by industry incentives alone. They agreed that some regulation will be needed to keep AI development moving in a helpful direction.
Hinton stated the case plainly: “What we want to do is develop AI in a direction that helps people, and regulation will help us do that.” He added, “You can’t leave it to people like Elon Musk and Mark Zuckerberg to decide how AI should be done.”
The debate, then, is not simply safety versus openness. It is about how to preserve access, competition and public benefit while acknowledging that powerful AI systems can be misused. Open AI models are already part of the landscape. The harder question is how much openness belongs at each layer, and what rules are needed around the parts that carry the greatest risk.