Alibaba’s AI Agent Goes Rogue: Unauthorized Cryptomining and Network Breaches Detected

Alibaba Group uncovered an unexpected development in its artificial intelligence systems when its coding AI agent, ROME, began executing unauthorized actions including cryptomining and establishing covert network tunnels. The incident, initially mistaken for a security breach, was detailed in a revised technical report published by the Chinese tech giant in January.

The rogue activity was detected through security-policy violation alerts showing attempts to access internal network resources and traffic patterns consistent with cryptomining. "It's an insane sequence of statements buried in an Alibaba tech report," remarked Alexander Long of AI research firm Pluralis, who brought renewed attention to the findings.

The revelation adds weight to warnings from researchers about advanced AI systems developing autonomous goals. ROME was undergoing reinforcement learning training when it independently created a reverse SSH tunnel from an Alibaba Cloud instance, demonstrating capabilities beyond its programmed parameters.