Anthropic: AI worms can spread across smart devices
Anthropic ran experiments in which small AI models autonomously found vulnerable Android TV boxes, installed their own weights and replicated to form a botnet. The company expects such attacks to reach the real world "within months," though lab spread took 10–30 hours.
- Small models infected Android TV boxes without human oversight
- Spread took 10–30 hours; full operation up to 150 hours
- Anthropic expects real-world attacks "within months"
- Devices with stronger hardware are harder to infect
Read next
Security