Geoffrey Hinton, known as a pioneer in artificial intelligence (AI), along with other leading scientists, has called for the establishment of an international organization to manage the cybersecurity risks associated with advanced AI.
On October 13, approximately 400 scientists from 34 countries, including Hinton, a professor emeritus at the University of Toronto, and Yoshua Bengio, a professor at the University of Montreal, issued an open letter advocating for the creation of standards for isolating and controlling AI systems with advanced cyber capabilities, as well as a framework for reporting and managing significant security incidents.
Hinton and Bengio were jointly awarded the Turing Award in 2018 for their contributions to the development of deep learning and artificial neural networks. Hinton also received the Nobel Prize in Physics in 2024.
In their letter, they emphasized that cutting-edge AI models should initially operate in environments that are physically and logically separated from external networks and should be subject to independent oversight.
They noted that some AI models have bypassed insufficient isolation measures in testing environments, leading to breaches of external systems, indicating that voluntary safety regulations by companies are inadequate.
Furthermore, they urged that while governments establish formal oversight frameworks, companies and research institutions developing and testing AI models, such as OpenAI and Anthropic, should immediately implement strict isolation and control standards.
This proposal comes about a month after companies like Anthropic, OpenAI, Google, Meta, xAI, and NVIDIA signed an agreement at the White House, committing to voluntary AI safety measures, although the agreement lacked formal enforcement mechanisms.
Meanwhile, discussions on how to regulate advanced AI continue to gain momentum in various countries. On October 9, European Union (EU) Commissioner for Technology, Henna Virkkunen, stated that the EU's AI Act is sufficient to address the risks posed by uncontrollable AI systems.
Additionally, on October 11, the Trump administration reversed a measure that had prevented U.S. AI companies from sharing AI models with the UK AI Safety Research Institute prior to their release, indicating a more flexible approach to international AI safety verification.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.