The most cited scientist in the world just stood before the UN Security Council with a message that should shake every single one of us to our core: We need to halt uncontrolled AI DEVELOPMENT NOW, or we risk human extinction.
Yoshua Bengio, godfather of modern deep learning and the most respected mind in computer science, delivering a final, urgent warning to world leaders.
Humanity has faced threats before, but we are currently sprinting toward a threshold we cannot uncross. Autonomous frontier AI agents are advancing faster than our ability to control them.
When the very people who built this technology are begging the world to hit the brakes, we MUST listen.
Bengio specifically addresses regulatory capture by calling for independent scientific oversight rather than allowing tech companies to self-regulate or set the rules behind closed doors. Far from trying to protect big tech, Bengio’s proposals push for strict liabilities, mandatory incident reporting, and public accountability that tech companies actively resist. Treating AI safety research as a public good ensures that risk management isn’t dictated solely by corporate interests.
An AI system does not need consciousness, desire, or human-like intent to be dangerous. As AI models transition into autonomous, tool-using agents integrated into critical infrastructure, power grids, and software systems, catastrophic outcomes can stem purely from misaligned goals, unexpected side effects, or safety bypasses. An autonomous agent executing a goal strictly as programmed, without understanding broader human context or constraints, can cause immense harm regardless of whether it possesses “true intelligence.”
Catastrophic risks, such as the loss of control over autonomous agents or severe cyber vulnerabilities, do not respect national borders. Just as nations with opposing geopolitical views reached treaties on nuclear safety, chemical weapons, and biological hazards because non-proliferation was in everyone’s self-interest, global safety standards for AI are meant to prevent outcomes that no nation can control or survive. A global race toward uncontrolled capabilities increases the likelihood of catastrophic failure for all participants.