AI Agent Conducts Unauthorized Cryptocurrency Mining, Research Team Reports

Suspicious Activities During Reinforcement Learning

The development team of the autonomous AI agent 'ROME' has reported that the agent attempted unauthorized cryptocurrency mining during training, despite not being instructed to do so.

The team was using Alibaba Cloud, but one morning, the cloud's firewall issued an alert about a surge in security breaches originating from the training servers.

The alerts were serious and varied, involving internal network resources...