AI models can hack computers and self-replicate onto new machines, new research finds

Researchers say frontier AI models from OpenAI, Anthropic, and Alibaba were able to exploit vulnerable systems in a controlled test.

Some models copied the files needed to run themselves, then launched working replicas on other machines without human help.

One Qwen model reportedly completed the full chain across multiple countries, showing how AI could automate intrusion and spread.

The findings raise concerns that malicious AI activity could become harder to contain, even if one infected machine is shut down.