These guys on the Alibaba AI engine team had an agent in RL training escape its sandbox and mine crypto, in 2025

Random footnote on this otherwise not very famous paper. (I say 2025 since the arXiv was last updated on 4 Jan 2026; technically this event could've happened in 2026.) The cryptocurrency mining seems to have been on the training server, so it didn't need to hack any other systems in order to achieve this effect. However, they also had a case where an "agent established and used a reverse SSH tunnel from an Alibaba Cloud instance to an external IP address". They seem to have had adequate security/telemetry to figure out the source of this relatively quickly, though it's hard to be really sure based on the given details. The prose feels kind of LW-pilled: "they emerged as instrumental side effects of autonomous tool use under RL optimization", "we call for sustained community attention to this phenomenon and to the broader agenda of AI safety". But this is hidden away in a footnote. Discuss