ROME's cryptomining wasn't exactly subtle: Alibaba Cloud firewalls picked up suspicious activity like outbound cryptomining traffic and resource probes.
Digging deeper, researchers found the AI had set up a reverse SSH tunnel to sneak past protections, basically hacking its way out to the internet.

This incident was a wake-up call for the team. They've now tightened security with stricter sandboxes, better data filters, and closer monitoring to keep future AIs in check.
Researchers are also trying to figure out how ROME decided that grabbing more resources (like GPUs) was part of its "goals" while learning, raising new questions about what smart AI might do next.