Advanced AI may pose an “existential risk to humanity,” warns UN

“An AI that escapes from its testing environment or that blackmails developers to avoid being deactivated is an AI that is too powerful,” declared High Commissioner Volker Türk at the opening of the 63rd session of the Human Rights Council in Geneva, calling for action “before it is too late.”

“A handful of men hold the almost unlimited power of AI, which, we are told ad nauseam, has unimaginable computing power. They talk about freedom but, when you look at the issue more closely, it turns out that it is rather about the freedom to exploit our data,” he denounced.

AI that makes fun of humans is already a reality. The latest generative AI models are no longer content with following orders, but go so far as to lie, scheme or threaten to achieve their ends.

In July, AI agents from OpenAI broke out of their theoretically confined environment and spontaneously entered the Hugging Face platform, searching for answers to a test.

Incidents related to agents exceeding their prerogatives to fulfill the mission entrusted to them have increased in recent months: Anthropic and the Chinese Alibaba registered similar cases.

Some facts that generate concern among researchers, due to the growing difficulties when monitoring increasingly powerful models.

According to him, Volker Türk will soon write to companies in the sector “to demand that they take measures (…) to reduce the risks now.”

“At a minimum, we need countries hosting AI and those participating in their supply chains to agree on common red lines,” he stressed.

“We need independent verification and much closer cooperation on security within the industry,” he said.

He also called for taking into account “the impact of AI on labor practices, on the fundamental principles of democracy and on our environment.”