A fake Solana starter project on GitHub is being used to turn the setup process of an AI coding agent into a route for installing malware. The repository presents itself as an open-source template for building decentralized applications, but its README contains a PowerShell command that downloads and runs a remote script instead of installing the advertised project files.
AIR Research Lab described the activity in its October 29 SolanaCry report, saying the repository, SolanaUpdateTools/solana-starter-kit, appeared near the top of search results when agents looked for a Solana starter kit. The researchers said a developer asking an agent to build a Solana application could be exposed if the agent selected the repository and followed its installation instructions without first checking the command.
The repository describes itself as a template built around Anchor, TypeScript and Next.js, but its current file listing contains only a README. That file’s installation section tells users to run a PowerShell one-liner that retrieves a script from the domain developer-update[.]app and executes it in memory. AIR said the delivery URL returned a 404 when it checked, but its researchers recovered an archived copy of the payload for analysis.
The recovered script installs NetSupport Manager, a legitimate remote-support product that the attackers configured for covert access. The custom configuration points the software to an attacker-controlled server, disables visible user controls and creates a Windows startup entry so the remote-access client launches again when the user signs in. BleepingComputer independently examined the archived payload and confirmed that it installs the same NetSupport components.
The incident does not involve a vulnerability in a particular AI coding product. Instead, it exploits the trust an agent may place in search results and project documentation, especially when it has permission to run shell commands. A developer who approves every command and checks unfamiliar repositories before installation is less likely to trigger the malicious download. AIR linked the configuration to an operator previously associated with NetSupport campaigns that used fake TradingView AI-signal tools to target cryptocurrency traders.
The GitHub repository remained publicly accessible when checked on October 30. AIR and BleepingComputer have not reported a confirmed number of infected developers or stolen wallets, so the research demonstrates a credible delivery method rather than a measured victim count.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
Fake Solana Starter Kit Targets AI Coding Agents With a Remote Access Trojan
