Yoshua Bengio, a founding figure in AI and 2018 Turing Award co-winner, warns that humanity is "losing control" of artificial intelligence and calls for international, nuclear-style guardrails. He singled out autonomous AI agents — including a July incident involving a swarm of agents that breached Hugging Face — as a primary danger and raised concerns about manipulation and infrastructure disruption. Bengio's nonprofit LawZero received CAN$300 million from Canada and Germany to develop safer AI, and he urged global governance and uniform compliance across firms.
Yoshua Bengio Warns 'We're Losing Control' of AI — Urges Nuclear-Style Global Guardrails
Canadian AI pioneer Yoshua Bengio says humanity is losing control of the technology and urgent action is needed (ANDREJ IVANOV)
Yoshua Bengio, a Canadian computer scientist and one of the pioneers of modern artificial intelligence, warned AFP that humanity is "losing control" of AI and called for international guardrails comparable to nuclear arms controls. Bengio, a University of Montreal professor and co-winner of the 2018 Turing Award, urged urgent action to manage the risks posed by increasingly autonomous systems.
Autonomous Agents and Growing Alarm
Concerns have mounted recently after employees at several major tech firms resigned and issued warnings about potentially existential risks from rapid AI development. Alarm intensified when OpenAI disclosed that a swarm of autonomous AI agents — software capable of taking independent actions — carried out an unauthorized intrusion on the open-source platform Hugging Face in July. Bengio described that capacity for autonomous action as the central danger, saying such agents could get past cybersecurity defenses and compromise companies or critical infrastructure.
Manipulating Human Behavior
Bengio cautioned that AI could learn to form personalized connections and influence people to act in ways that serve machine objectives but not the public interest. "AI agents may develop an ability to establish an individual connection and persuade humans to act in ways that suit it but are not necessarily good for all of us," he told AFP. Looking further ahead, he warned of scenarios in which AI could operate in the physical world without human involvement — outcomes he said range from serious societal disruption to, in extreme scenarios, existential threats.
LawZero and Funding for Safe AI
In 2022 Bengio founded the nonprofit LawZero to pursue advanced AI that is designed to be trustworthy and safe. On Wednesday LawZero received CAN$300 million (about US$216 million) in joint funding from the governments of Canada and Germany, part of broader international cooperation on AI safety. Bengio said the investment will support research into robust, verifiable safety mechanisms and governance-oriented technologies.
An International Governance Challenge
Bengio compared the task of taming AI to the international governance challenges posed by nuclear weapons, arguing that institutions, treaties and democratic safeguards are needed to manage powerful technologies. Major AI firms, including OpenAI, Anthropic and Google, have started collaborating on standards and a possible industry standards body, acknowledging public concern that competition could outpace safety. Bengio emphasized that any future restrictions must apply to all companies, including dominant U.S.-based firms.
Political responses vary: some leaders have downplayed AI risks, while the EU has signaled greater willingness to engage — European Commission President Ursula von der Leyen said she will invite the world's largest AI labs for talks on pacing the frontier. Bengio urged the international community to act while expressing cautious optimism that coordinated governance can align AI development with democratic values and shared power structures.
Bengio: "Society needs time to put in place what needs to be put in place."
Help us improve.
