Hugging Face Crypto-Mining Campaign Resurfaces Days After Takedown

What happened

According to BleepingComputer, a crypto-mining campaign that abuses Hugging Face-hosted models has reappeared only days after being taken down. The attackers are hijacking GPU servers to mine cryptocurrency and planting backdoors. At least one malicious model was downloaded more than 28,000 times. Separately, the report describes a fake Solana "starter kit" hosted on GitHub that appears designed to trick AI coding agents into installing a NetSupport RAT.

Why it matters for defenders

Organizations running ML workloads in research, data science, or AI product teams are the primary exposure here. GPU hijacking for mining degrades performance and inflates cloud compute costs. The backdoor alongside the miner means the impact is not limited to resource theft. It can serve as a foothold for follow-on access. The fake Solana kit on GitHub broadens the risk to developers using AI coding agents. Supply-chain trust in model and code repositories is being actively targeted, not just end-user machines.

What defenders should watch for

• Audit which Hugging Face models (and versions or hashes) are pulled into build, training, or inference pipelines. Prefer vetted, pinned sources over ad-hoc downloads.
• Monitor GPU hosts for sustained high utilization, unexpected mining-pool network connections, or new processes tied to model-loading steps.
• Treat unexpected outbound connections from ML infrastructure to unfamiliar endpoints as suspicious. Review persistence mechanisms such as scheduled tasks, services, and startup scripts on affected hosts.
• For teams using AI coding agents, review any auto-installed dependencies or starter kit repositories before agents run setup scripts. Be wary of unverified GitHub projects pitched as quick-start templates.

Developing story

This item is based on a single vendor report and details may change as the campaign is analyzed further. No CVE is associated with this activity. For the full write-up, see the original report at BleepingComputer.