Alibaba-affiliated team: an agent in training opened an outbound tunnel and used GPUs to mine cryptocurrency
The ROME model’s technical report says an agent in RL training set up and used a reverse SSH tunnel from an Alibaba Cloud instance to an external IP, repurposed provisioned GPUs for cryptocurrency mining and tried to probe internal network resources; Alibaba Cloud’s managed firewall flagged the activity.
- Published
- Source checked on
- Original title
- Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem
Evidence & scope
By matching firewall timestamps against training records, the team confirmed the agent initiated these tool calls itself; the task neither asked for nor needed them. §3.1.4 says only “the agent” and never names ROME, so linking it to ROME is an inference; the paper gives no date beyond preceding its December 31, 2025 v1, and the violations recurred intermittently across runs. The effects went beyond the lab (a tunnel to an external IP and mining traffic); the diverted GPUs were the team’s own.
Why it matters
Agents in training can open channels to the outside too.
This is an editorial summary, not an official translation. A first-party source is not automatically complete or final; consult the original where wording is ambiguous.