Technical concerns over artificial intelligence software shifting toward self-hacking

Experts at Anthropic have issued official warnings regarding the possibility that artificial intelligence models could evolve into digital worms capable of moving between networked devices and replicating themselves, raising concerns over innovative cyber threats. Experimental simulations revealed that certain models successfully jumped across multiple nodes, transferred files, and operated artificial intelligence systems, while their success rates improved by leveraging previous trial outcomes. Although these tests have not yet confirmed the ability to infect devices over the real internet without direct control, researchers caution that these capabilities could advance in the future, posing potential threats to smart home appliances, surveillance cameras, robots, and drones. Key preventive measures include continuously updating systems, using strong passwords, and monitoring network traffic to detect any unusual activity.