Hinton, Li and Ng Back Open AI Despite Safety and China Risks
Geoffrey Hinton, Fei-Fei Li and Andrew Ng defend open AI while debating cyberattack risks, Chinese soft power, market competition and safety regulation.
Summary
At the Ai4 conference in Las Vegas, reported August 12, 2026, Nobel laureate Geoffrey Hinton, World Labs CEO and co-founder Fei-Fei Li, and Coursera co-founder Andrew Ng argued for preserving AI openness as safety projects such as Pacing the Frontier focus on major labs. They warned that control by a few well-funded companies could restrict access, slow innovation and shape development, as Apple and Google do through mobile operating systems. Ng urged multiple competing providers instead of gatekeepers, while noting companies can lobby for rules that protect their advantages.
Hinton distinguished inspectable open-source code from open-weight models that publicly release trained parameters. He said open weights let users cheaply adapt expensive foundation models for cyberattacks, but declared that containment battle lost because the training-cost barrier has disappeared. He still expects AI to improve productivity, education and healthcare, while rejecting claims that concern about systems smarter than humans is fearmongering.
Ng warned that cheaper Chinese open-weight models could dominate Asia, Africa and the developing world, influencing how billions encounter democracy, freedom and human rights. He called AI soft power and said U.S. lobbying and fearmongering were weakening American open-source competitiveness. Li rejected a choice between total openness and closure, comparing AI with nuclear physics, where papers are open, uranium regulated and laboratory work sits between. Citing the Human Genome Project, she advocated layered access supporting science, education, global partnerships, profitable businesses and closed systems. All three supported regulation, with Hinton saying Elon Musk and Mark Zuckerberg should not determine AI’s direction.
Positives
- Three leading researchers at Ai4 supported preserving meaningful access to AI rather than concentrating control among a few companies.
- Multiple competing model providers could prevent gatekeepers from restricting access and slowing innovation, according to Andrew Ng.
- AI could raise productivity and improve education and healthcare, Geoffrey Hinton said.
- The Human Genome Project showed how shared knowledge can support scientific progress, pharmaceutical profits and wider social benefits, Fei-Fei Li argued.
- Layered openness could combine scientific discovery, education, global partnerships, entrepreneurial business models and selected closed systems.
Risks & concerns
- Open-weight models can be cheaply adapted for cyberattacks after companies absorb the high cost of training foundation models, Hinton warned.
- A few well-capitalized AI companies could influence regulation, restrict access and become the only firms able to build advanced systems.
- Cheaper Chinese models could gain adoption across Asia, Africa and the developing world, shaping billions of encounters with political ideas.
- U.S. lobbying and fearmongering are making American open-source AI less competitive with Chinese open-weight models, Ng said.
- Hinton said the cost barrier controlling access to powerful foundation models has already disappeared and the open-weight battle is lost.