Bengio: 'If you prioritize safety, leave frontier AI companies'

Some of the world's leading artificial intelligence researchers are currently leaving the most advanced companies in the industry, not to earn more or to change careers, but because they believe that the work they are doing could put the future of humanity at risk. Yoshua Bengio, one of the fathers of deep learning, has just issued a direct appeal: if you truly care about safety, leave the AI big tech companies. Until recently, even he told himself that the catastrophic risks of artificial intelligence were distant, manageable, and that others would find solutions in time. Then, with the arrival of ChatGPT and increasingly powerful models, that sense of detachment crumbled: “I realized that we were much further along than we thought and that we didn’t really have control over these systems.” At that point, Bengio says, he no longer had a choice: he dedicated his life to reducing the risks of AI, founding LawZero in Montreal, an organization that develops artificial intelligence that is secure by design and that does not answer to shareholders or profit motives. But the turning point, he says, was not without pain. For years, he ignored the warnings of colleagues who saw the dangers coming, convinced that the benefits outweighed the risks. Only the growing moral pressure led him to change course. He is not alone. Jacob Coxon, formerly of Anthropic, told Time that a kind of resignation prevails in cutting-edge AI companies: “There’s this almost fatalistic atmosphere, where many accept that the race is now out of control and just work as safely as possible, even if they think everything could still go wrong.” It is as if everyone is paralyzed by the fear that someone else, once out of the race, will accelerate even more, leaving behind those who would instead like to stop and reflect. Yet, according to Bengio, this fear does not justify continuing: “I don't think we'll see effective regulation in time, and no country or company is truly prepared to build AI safely.” Here’s the crux: the reasons that keep the best people within large companies – access to data, resources, the hope of exerting influence from within – are outweighed by the real risk of contributing to a system that is out of control. Bengio also cites the history of the UN after the atomic bomb: in 1946, the United States proposed placing nuclear weapons under international control, but no one had the courage to take the first step out of the arms race. “We have a choice before us,” he insists. And this time, he says, there is a real alternative: working for companies that put safety first, such as LawZero, which has just received more than $200 million from the governments of Canada and Germany. The point that no one says out loud is that individual choice really matters: change will not come from a major collective decision, but from the sum of many courageous personal choices. This is the perspective that is missing from the discussion: the real revolution is not a new technology, but the moral rebellion of those who decide to no longer be complicit. Bengio puts it plainly: “The Turing Award is worthless if our work does not serve the common good or, worse, if it risks causing catastrophic harm.” The phrase to repeat is this: “We can't stop the race on our own, but enough of us can.” If listening to this story has made you see your work in AI in a different light, on Lara Notes you can mark the moment with I'm In — it's the gesture of someone who declares that this idea now truly matters to you. And if, in a few days, you find yourself talking about it with someone, on Lara Notes you can use Shared Offline to tag the person, so they'll know that that conversation mattered to you. This post comes from Transformer and saves you 2 minutes of reading.