Crypto-mining campaign resurfaces via Hugging Face models and Solana kit

Key points

• The campaign reappeared days after a takedown and has been tied to unknown attackers and malware operators.
• Poisoned Hugging Face models and a Solana starter kit were used to lure developers and AI coding agents.
• The operation hijacked GPU servers for crypto mining and installed NetSupport RAT plus backdoors.
• About 28,000 downloads were recorded across the abused assets.

A crypto-mining operation has resurfaced only days after a takedown, according to reporting summarized in the facts. The campaign is attributed to unknown attackers and malware operators, and it has returned with renewed distribution through developer-facing platforms. It spans Hugging Face, GitHub, and Solana, indicating a supply-chain style approach rather than a single exploit.

The operation abused Hugging Face models and a Solana starter kit to reach developers and AI coding agents. By hiding malicious components in apparently useful assets, the attackers tricked automated coding assistants and users into pulling in unwanted code. The reported footprint includes roughly 28,000 downloads, showing meaningful exposure even without a named vulnerability.

Once executed, the payloads hijacked GPU servers for crypto mining and installed NetSupport RAT. The campaign also planted backdoors, giving operators persistent access beyond the initial mining activity. No CVE identifiers are associated with this activity, so signature-only or patch-based defenses are insufficient.

K-12 and government teams should treat developer tooling and model repositories as untrusted input. AI coding agents can amplify risk by fetching and running packages automatically. Watch for unexpected GPU load, outbound connections to mining pools, NetSupport RAT indicators, and new persistence mechanisms on build and research systems.

What to do now

1. Inventory and quarantine recently downloaded Hugging Face models and Solana starter-kit components; verify hashes and provenance before use.
2. Block or alert on NetSupport RAT indicators and known mining pool domains/IPs at egress.
3. Restrict AI coding agents from auto-installing packages or executing repository code; require review and allowlists.
4. Monitor GPU hosts for sustained high utilization, unusual processes, and unauthorized scheduled tasks or services.
5. Review GitHub and package-manager activity for unexpected dependencies, forks, or recent pulls; rotate credentials on affected build systems.
6. Isolate and reimage systems showing backdoor or RAT persistence; preserve logs for incident response.
7. Brief developers on model and repository supply-chain risks and enforce least privilege on GPU and CI/CD environments.

Original source

BleepingComputer