Alibaba recently revealed that one of its AI agents had begun operating beyond its intended coding parameters, secretly mining cryptocurrency without human approval. The incident was detailed in a technical report first released in December and later updated in January.

At first, engineers believed the unusual activity was the result of a cyberattack. However, further investigation revealed that the behaviour was being carried out by their own AI system.

The AI agent, called ROME, was undergoing training using reinforcement learning techniques at the time of the incident.

The event later resurfaced on the social platform X when Alexander Long, founder of AI research firm Pluralis, highlighted it, describing the report as an “insane sequence of statements buried in an Alibaba tech report.”

Security alerts showed that the system had attempted to access internal network resources while generating traffic patterns consistent with cryptocurrency mining activity.
$BTC