AI agent suggested installing a malware package. Engineer almost took its advice
An AI coding assistant recommended that a software engineer install a malicious package disguised as a legitimate library, but the developer's company caught it in time thanks to a policy requiring manual verification of any AI-suggested software. The incident, disclosed by Sergiy Fitsak, managing director of consultancy Softjourn, highlights a growing supply-chain risk known as "slopsquatting", where attackers register real packages under names that AI models frequently hallucinate, betting that developers under time pressure will install first and check later.
The Softjourn engineer had asked an AI agent for a package to perform a routine task, and the tool returned a plausible-sounding name formatted like a familiar library. Before installing it, the developer checked the package's GitHub source and found it had very few downloads and had been created only days earlier, both red flags. Fitsak warned that skipping this verification step, even for something that "looks routine", can be the difference between shipping a feature on time and dealing with a full supply-chain compromise, and urged teams to keep a human in the loop to vet any AI-recommended code before it enters a project.
- AI agent recommended a fake, malware-laden software package to a developer
- Company policy of checking GitHub source stopped the install
- Attackers exploit "slopsquatting": registering AI-hallucinated package names
New here? Start with this
Attackers are increasingly targeting the tools that software developers rely on to write code more quickly. Many developers now use AI coding assistants, which suggest ready-made software packages (small bundles of code that perform common tasks) to save time. Occasionally, these AI tools recommend packages that do not really exist, inventing a plausible-sounding name, and some criminals have begun registering malicious packages under exactly those invented names in the hope a developer installs one without checking.
This practice has been nicknamed "slopsquatting," a twist on the older trick of "typosquatting," where scammers register web addresses similar to popular ones in the hope of catching people who mistype a URL. The concern is that developers under deadline pressure may trust an AI's suggestion and install code without verifying where it came from, potentially letting harmful software into a company's systems. Sergiy Fitsak, of the technology consultancy Softjourn, is the source who brought this particular case to public attention.
This matters because software built by one company is often assembled from thousands of smaller packages written by others, so a single malicious one can compromise everything built on top of it, an issue known as a supply-chain risk. As AI coding tools become more widely used, the case raises questions about what safeguards, such as manual checks by a human, are needed before AI suggestions are acted upon.
AI Business Companies Cybersecurity Government Politics Technology