The latest controversy regarding potentially dangerous uses of artificial intelligence (AI) focuses on the technology sector and its application to military weaponry.
As reported by Forbes, the company Anthropic, responsible for the AI assistant Claude, has banned initiatives that -possibly- intended to use its model to create biological weapons, the company announced on Thursday, revealing the findings of an investigation just days after "AI experts warned about the potential of increasingly powerful digital assistants to cause human extinction."
The company stated in a report that it detected and blocked these misuse cases during this year, although it could not determine whether the studies had legitimate or malicious purposes, so it chose to block them due to the risk that their models could contribute to the development of dangerous pathogens.
"While assessments are useful as they provide evidence of capability, they cannot definitively prove that such capability would ever be used to develop biological weapons in the real world," the company explained in the document.
So far, the California corporation denied having any knowledge of any private company publicly sharing "evidence of possible misuse of their platforms for the development of biological weapons."
Anthropic also included in another statement an extensive list of cases detected in the last eight months where their AI models had been misused, including Claude, in China, Russia, and Yemen.
This report comes amid growing concerns from experts in the tech sector about the dual use of AI, as models capable of assisting in biological research can also be used for dangerous activities, as well as warnings about the risks associated with the increasing autonomy of these models.
The tech company claimed to have developed detection systems to identify and block any uses that may pose a risk in their AI models.
Yesterday, it was reported that a researcher from Anthropic is leaving the artificial intelligence industry out of fear that the lab and its competitors are racing to build systems that they will not be able to control, a sign of the growing concern for security within leading AI companies.