SMALL AI MODELS NEARING SELF-REPLICATION CAPABILITY

SAN FRANCISCO, OCTOBER 13 /SRNA/ - Tests by Anthropic, an artificial intelligence /AI/ research and safety company, suggest that smaller AI models could within months become capable of replicating themselves onto devices such as smart microwave ovens and Android TV boxes.

In its experiments, Anthropic examined whether AI models could identify vulnerable devices, copy themselves onto them and continue spreading without human assistance.

The company found that smaller models were capable of carrying out many of the steps required for replication.

However, it found that similar AI models still had difficulty completing the entire sequence.

In simplified tests in which certain security measures were removed, some more powerful AI models were able to complete the replication process.

"Self-replicating models are difficult to shut down because there is no single point at which they can be completely disabled and, unlike non-AI malware, they can identify previously unknown devices and target their processors," Anthropic warned.

The company focused on devices such as Android TV boxes, smart home appliances and other internet-connected devices because they generally have weaker security protections.