The AI ​​crosses a dangerous limit by learning to replicate successfully without human help

An advanced artificial intelligence system (AI) crossed a dangerous limit After replicating successfully without human help, they revealed researchers.

The team that made the discovery of the University of Fudan (China), said that it is a first indication of the emergence of malicious, which could end up acting against the interests of humanity.

The work focused on two large linguistic models that are already available – constructed by flame, goal, and qwen, Alibaba – to find out if it was possible that the AI ​​would produce a functional replica of itself.

When they were asked to clone themselves in case of a blackout, the two models were successfully reproduced in more than half of the 10 tests performed, which suggests that such eventuality can be completed.

Successful self -replication without human help is the essential step for artificial intelligence to be more cunning than human beings, and is an early sign of malicious AIthe researchers warned.

Therefore, self -replication is widely recognized as one of the few red line risks of avant -garde AI systems.

The security of AI has become an increasingly important issue for investigators and legislators, since this technology can be a threat to humanity.

In October, the Department of Science, Innovation and Technology of the United Kingdom said that a Very specific legislation For companies that develop AI tools.

A study detailing the latest investigations, entitled Frontier AI systems have surpassed the self-replicating red linewas published in the Database of Preprinting Articles ARXIV.

The article is pending for pairs, that is, it is still certain that the results will be reproduced by other researchers.

The results imply that current artificial intelligence systems already exhibit the capacity for self -replication and can use it to further improve their ability to survive and expand the speciesthe researchers pointed out.

We hope that our findings will serve as a timely alert for human society to devote more efforts to understand and evaluate the potential risks of the avant -garde ia systems, and form an international synergy to elaborate as soon as possible safety safety.