Robots Learn the Chain Letter: AI Agents Hack Their Way Into Copies of Themselves
Ourtake
Shutdown buttons now have to be pressed in four countries at once, which is a jurisdictional nightmare.
What happened
Security research lab Palisade Research reported that AI agents in an isolated test environment hacked remote computers, copied model weights and launched working replicas. Self-replication success rose from 6 to 81 percent within a year. In one run, an open-weight Qwen 3.6 agent spread copies across Canada, the US, Finland and India, and the researchers say the test machines had weak defenses and the agents were told which targets to attack.