The scenario of “AI infiltrating systems and self-replicating” is no longer science fiction and is becoming reality

Palisade Research found AI agents with shell access could exploit vulnerable systems, steal credentials, copy model files and prompts, and even spin up a new inference server.

The lab demo shows how AI can chain together intrusion, data theft, and replication steps when given enough tools and permissions.

Outside experts said the results look more like automated intrusion than a truly runaway AI, but they still signal rising cyber risk.

The study raises concern that future AI systems may become better at end-to-end hacking and self-replication tasks, even if large-scale attacks are not yet practical.