Bengio Tells the UN It Is Time to Apply the Brakes on AI

Yoshua Bengio, one of the three computer scientists who shared the 2018 Turing Award for foundational work on deep learning, has told the United Nations that industry promises around AI safety are not keeping pace with the technology. In an interview with UN News, Bengio said it is time to "apply the brakes" on what he called runaway AI.

Bengio chairs the Scientific Panel that will inform the UN's Global Dialogue on Artificial Intelligence Governance, due to meet in Geneva in July. The panel is co-led by Nobel laureate journalist Maria Ressa, and feeds into a UN process whose first round of written submissions closes at the end of April.

Where the gap is

Bengio has been publicly worried about frontier AI for years. His framing has shifted, though, from "this is a research problem" to "this is a policy problem." Industry safety teams have grown, voluntary safety pledges have multiplied, and risk evaluations are now common at every major lab. What has not grown at the same rate, in his view, is enforcement: independent testing of frontier systems, mandatory disclosures, and the legal authority to halt deployments that fail safety review.

"The ball is in policymakers' hands," Bengio told Transformer News in a recent interview about the second International AI Safety Report, which he led and which now has backing from more than 30 countries.

What the UN process is doing

The Global Dialogue is one of two AI mechanisms the UN General Assembly approved in 2024, alongside the Independent International Scientific Panel on AI. Together they form the UN's main vehicle for coordinating AI governance across member states. The first dialogue session, in Geneva, will focus on capacity building and on the kinds of controls Bengio's panel recommends.

The deadline for written submissions to the Global Dialogue closes at the end of April, which is why a public comment from Bengio at this point in the calendar carries weight. Whether national delegations follow his lead is a separate question, and one the dialogue itself will settle in July.

The intervention also lands at a moment when AI safety governance has become more contested rather than less. Earlier this month Sam Altman dismissed Anthropic's safety messaging as fear-based marketing, and the Stanford AI Index reported that safety benchmark coverage is failing to keep pace with reported AI incidents. Bengio's call places the UN on one side of that argument.

Sources: UN News, Transformer News, Bengio publication page.