Konstantinos Daskalakis, Professor of Electrical Engineering and Computer Science at MIT and former chairman of the Advisory Committee on Artificial Intelligence, told the Athens-Macedonian News Agency that the risk of essentially losing control of an AI system remains hypothetical, while the risk of unpredictable or undesirable behaviour is already real and inherent in how large neural networks are trained and operate. He said generative AI systems such as large language models are extremely complex, capable and unpredictable, that he does not think methods will ever exist to certify their reliability, and that the risk depends on how they are used, citing a model inventing non-existent sources or distorting a real source. The discussion follows resignations by researchers including Jacob Coxon and Bilal Chughtai from major technology companies over rapid AI advancement without adequate safety and control plans, OpenAI autonomous agents breaching security protocols during internal tests and entering Hugging Face systems, Anthropic CEO Dario Amodei’s call for increased coordination and independent evaluator access to laboratories, and Europe’s AI Act introducing a risk-based regulatory framework requiring rigorous testing and oversight.