Three AI pioneers make the case for open models, even as safety fears grow

As efforts like Pacing the Frontier push major AI labs toward safety-focused coordination, open-weight models have become an industry sore spot — freely distributed and hard to control, they make some labs nervous. But at the Ai4 conference in Las Vegas last week, three of the field’s most respected researchers — Nobel Prize winner Geoffrey Hinton, World Labs CEO Fei-Fei Li, and Coursera co-founder Andrew Ng — each made a case for keeping AI open, even as they disagreed on the details.

Their shared worry: letting a handful of major AI companies control the pace of progress, the way Apple and Google control mobile operating systems. Ng argued for maintaining multiple competing providers rather than letting a few firms dominate. “I don’t want there to be gatekeepers,” Ng said. “That limits how all of us can access AI.”

Hinton took a more cautious view, drawing a sharp line between open source software — where the underlying code can be inspected — and open-weight models, which release only a trained model’s parameters. He said he had opposed releasing open weights because it makes it cheap for bad actors to repurpose expensive foundation models for harmful uses like cyberattacks, though he acknowledged that ship has sailed: open-weight models are now a permanent fixture, and the barrier they once posed has disappeared.

Ng reframed the debate around competitiveness rather than risk alone, warning that if cheaper open-weight models from China gain wide adoption across the developing world, they could shape how billions of people encounter ideas about democracy and human rights.

Li rejected the binary altogether, arguing that openness isn’t all-or-nothing — different layers of an AI ecosystem, like nuclear physics research versus uranium itself, can carry different levels of openness. She pointed to the Human Genome Project as a model where open scientific infrastructure let both researchers and companies build value.

Despite their differences, all three agreed that some regulation will be necessary to keep AI development on track.

Why it matters

The debate signals that “open vs. closed” is splintering into a more granular argument over which layers of AI — weights, code, safety research — deserve openness, with global competitiveness now as central a concern as safety.