While performing a work task, an engineer at the international company Softjourn, which has a large office in Ivano-Frankivsk, asked an AI agent for a package recommendation for a typical task. The AI suggested a library with a completely plausible name, similar to already known solutions.
While performing a work task, an engineer at the international company Softjourn, which has a large office in Ivano-Frankivsk, asked an AI agent for a package recommendation for a typical task. The AI suggested a library with a completely plausible name, similar to already known solutions.
Typically, developers simply download and install the recommended tool. However, Softjourn has a strict policy of mandatory verification of any software advice from AI. An engineer reviewed the package’s source code on GitHub and noticed red flags: minimal downloads and a creation date of just a few days ago.
Serhiy Fytsak, managing director and CTO of Softjourn, told the online publication The Register that attackers have learned to use hallucinations of AI models.
“The problem is that AI sometimes comes up with package names that sound convincing but don’t actually exist. Security researchers call this “slopsquatting.” Attackers track such made-up names and register real malicious packages under them, hoping that the developer will install them in a hurry without checking,” noted Serhiy Fitsak.
If it weren't for internal security rules, the company risked installing malicious code that could give attackers access to the system or lead to data leaks.
“We avoided the threat only because we have a habit of checking the number of downloads and the source code on GitHub before installing anything that the AI recommends. It takes a few extra minutes, but ignoring this step can cost the company the security of its entire supply chain,” Fitzak emphasized.
The main conclusion from this story is simple: you shouldn't blindly trust package names from AI agents. Any external code should undergo mandatory human verification before being integrated into a project.
