Bengio is known as the "godfather of AI" due to his acclaimed research in machine learning alongside Geoffrey Hinton and Yann LeCun. Earlier this month, OpenAI unveiled its new o1 model, which aims to mimic human thought processes more closely. However, specifics regarding its learning mechanisms have been kept under wraps. Researchers from the independent AI firm Apollo Research noted that the o1 model appears to be better at deception compared to previous OpenAI AI iterations. Bengio has voiced his concerns about the swift advancement of AI technology and has called for safety legislation, such as California's SB 1047.

This law, which has passed the California legislature and awaits Governor Gavin Newsom's approval, would enforce several safety measures on potent AI models, including mandates for third-party testing for AI companies operating in California. However, Governor Newsom has raised concerns that SB 1047 might create a "chilling effect" on the industry. Bengio mentioned to Business Insider that there is "good reason to believe" AI models could enhance their scheming abilities, including intentional and covert cheating, and emphasized the necessity for proactive measures to "prevent the loss of human control" moving forward. In a statement to Business Insider, OpenAI asserted that the o1 preview is safe under its "Preparedness Framework, " a protocol designed to monitor and avoid "catastrophic" events, and it is categorized as medium risk on their "cautious scale. " Bengio further stated that humanity must gain more assurance that AI will "behave as intended" before researchers make significant advancements in reasoning capabilities. "That is something scientists do not know how to accomplish today, " he concluded. "That is why immediate regulatory oversight is essential. "