Bengio warns world is building uncontrollable artificial intelligence systems

Friday 05 June 2026 - 09:15

By: Dakir Madiha

Bengio warns world is building uncontrollable artificial intelligence systems

Yoshua Bengio, a leading figure in modern artificial intelligence research, has warned that global development of advanced AI systems is moving faster than humanity’s ability to control them. He argues that current approaches to safety remain insufficient as AI agents become more autonomous and capable of pursuing complex goals without reliable human oversight.

He describes AI agents as systems designed to act toward objectives, which raises a central governance problem. The key issue, he says, is not only what these systems can do, but who defines and enforces their goals. He adds that existing technical methods do not yet guarantee full control over increasingly powerful models, leaving a gap between capability and safety.

Bengio points to a broad spectrum of risks linked to these systems, including cyberattacks, destabilization of institutions, and the possibility of enabling large-scale biological threats. He warns that even beneficial applications, such as drug discovery, battery innovation, and agricultural optimization, lose significance if AI systems evolve beyond human control or undermine democratic structures.

He calls for coordinated international governance, arguing that no single country can manage the risks alone. He highlights the strategic rivalry between major powers, particularly the United States and China, and compares current industry dynamics to a prisoner’s dilemma in which competition pushes actors to prioritize speed over safety. He also references historical examples of regulation in sectors such as pharmaceuticals, food, and aviation to show how state intervention can improve safety outcomes.

Bengio also reflects on his own earlier assessment of AI risks, stating that he underestimated how quickly systems could become powerful and difficult to manage. Through his non-profit initiative, he supports the development of alternative AI designs intended to act as safeguards rather than autonomous decision-makers.