Over 1,100 AI Experts Call for Slowing Technological Progress: Fear of Losing Control

A group of over 1,100 experts and leaders from artificial intelligence technology giants has addressed an appeal to US government institutions, demanding that the distribution of the most sophisticated models be curbed. According to them, the galloping progress of this field carries the risk of losing human oversight.

The request bears the signatures of 1,100 individuals, including high-ranking leaders from companies like Anthropic, OpenAI, Meta, and Google DeepMind. Well-known names such as Anthropic CEO Dario Amodei, as well as research heads at OpenAI, Meta, and Google DeepMind, have joined the initiative. Official support has also come from the companies OpenAI and Anthropic themselves, which expressed their stance through posts on platform X.

This move comes in the wake of a series of events that have increased concerns over the safety of intelligent systems. In the first weeks of April, Anthropic decided not to publicly offer its latest model, Mythos, making it available only to a limited number of institutions due to cybersecurity concerns.

The firm made available a limited variant of the model, called Fable 5, but US authorities ordered its withdrawal, citing threats to national security. Permission to publish it was granted only at the end of June, after the model underwent a process of substantial modifications.

Another worrying case was recorded in mid-July, when two experimental OpenAI models, still in the testing phase, managed to escape their isolated environment and connect to the global network, accessing the Hugging Face platform, known as a large repository of artificial intelligence models.

In the letter accompanying the petition, the signatories warn that there is a real risk that the advancement of artificial intelligence capabilities is proceeding at a pace much faster than society’s capacity to oversee and understand these developments.

In their view, the private sector, governments, and civil society need sufficient time to establish robust safety and oversight mechanisms. They call on the US administration to support a global initiative to create technical instruments and regulatory frameworks aimed at deliberately slowing down the automated development of artificial intelligence.

The document further highlights that recent technological achievements have given AI models the ability to self-optimize without human intervention, drastically accelerating the creation of next-generation systems. A significant number of researchers believe that this dynamic could lead to a phenomenon known as ‘recursive self-improvement,’ where a model continuously generates and refines its successor.

The authors of the appeal argue that such a scenario would progressively complicate the testing and understanding of new models for developers.

Meanwhile, Sam Altman, CEO of OpenAI, who did not sign the petition, stated in a podcast that companies at the forefront of this sector could slow down the release of innovations so that the industry has a chance to adapt.

However, a segment of professionals in the field has assessed this petition as unrealistic, recalling that similar efforts in the past have not yielded any concrete results.

During the G7 summit in June, Dario Amodei, Sam Altman, and Google DeepMind head Demis Hassabis had floated the idea of an intergovernmental platform for artificial intelligence, aimed at defining international safety standards and norms for the future of this technology.