What is SlopSquatting

1 minute read

What is SlopSquatting?

SlopSquatting is a portmanteau of AI Slop and Typosquatting. When coding with GenAI, it may unknowingly suggest a malicious library based on web search results. For example, attackers can upload their modules to public repositories like GitHub and describe them using plausible-sounding documentation or keywords to make them appear trustworthy. This is intended to trick GenAI into recommending these malicious packages in generated code – especially when the AI relies on popularity signals or star ratings.

A real-world analogy is typosquatting attacks in package managers (e.g., requests vs reqeusts in PyPI), but SlopSquatting goes further by relying on sloppy, automated decision-making from LLMs rather than human error.

How can we prevent it?

There are two major strategies to reduce risk:

1. CI/CD-Based Safeguards

Integrate security tools into your build pipeline to scan dependencies and detect supply chain risks. Examples include:

JFrog Xray, Snyk, or ChainGuard for vulnerability scanning
Terraform, CircleCI, or GitHub Actions for automation and enforcement
Software Bill of Materials (SBOM) generation and validation

These tools can detect known bad packages or unexpected transitive dependencies before your code hits production.

2. Prompt Engineering and Human-in-the-Loop

When using GenAI for code generation:

Include explicit instructions in prompts such as “only use well-known, secure libraries” or “cross-check modules against official documentation”.
Build a verification step into your workflow. For instance, ask the LLM to explain the purpose and origin of each module it suggests.

⚠️ Note: This method is still experimental – I haven’t confirmed if it consistently works across different LLMs.

Notable Real-World Example

In early 2024, researchers demonstrated that ChatGPT could be manipulated to recommend typosquatted or malicious npm/PyPI packages simply by seeding GitHub repositories with realistic-looking READMEs and keywords. This behavior is detailed in papers like:

“Are LLMs Safe Package Recommenders?“ (arXiv 2024)

GitHub threat reports from ReversingLabs and Checkmarx

These findings highlight how LLMs trained on public code and search results can unknowingly propagate security threats when used naively in software development.

What is SlopSquatting