A package can look normal, pass every CVE check and still run harmful code the moment you install it. That's because CVE lists (public, numbered records of security flaws) track honest mistakes, not code written on purpose to do harm.
Plus, these attacks come in many shapes. Some packages copy a popular name with one typo. Some come from a maintainer account an attacker took over. And some hide scripts that run on their own during install. From there, the damage can spread to your developers' laptops, your build systems and the software you ship.
So, you need a different kind of tool to catch them. Here are eight, with a quick summary and a table up front, then each tool's pros.
Next, here's a side-by-side view. Each row shows what a tool's own pages stress most, and most tools do more than one row can show.
| Tool | What it catches | Where it checks | Ecosystems |
| Aikido | Malware, typosquats, dependency confusion, compromised releases, install-script attacks, data theft, crypto miners | Install (laptops and CI), repos, containers, IDE, developer devices | npm and PyPI tools at install; Device Protection adds Maven, NuGet, Go, Ruby, Rust, PHP and extensions; repo scanning includes npm, PyPI, Maven, NuGet, Go and Rust |
| Socket | Known malware, typosquats, risky behavior like network calls and shell commands | Install (machines, CI, network), pull requests | Free Firewall: npm, Python, Rust; Enterprise adds Maven, Ruby, NuGet and more |
| Endor Labs | Malicious code, typosquats, dependency confusion, install-time scripts, hijacked maintainers | Install (developer machines, AI agent workspaces, CI) | npm, PyPI and other registries |
| Sonatype Firewall | Malicious packages, parts that break your rules, risky AI models and container layers | Point of download (repository, network edge, work computers) | npm, Maven, PyPI, NuGet, plus Hugging Face models and Docker images |
| JFrog Curation | Typosquats, hijacked packages, data theft, dependency confusion, too-new releases | Point of request in Artifactory, including AI agent requests | Packages, AI models and IDE extensions |
| Checkmarx | Known malware, suspicious behavior, typosquats, dependency confusion, hijacked maintainers | Package lists, compiled files, containers; pull request and build checks | npm, PyPI, Maven, RubyGems, NuGet, Go, Rust and more |
| Mend SCA | Typosquats, dependency confusion, data stealers, backdoors, protestware, crypto miners | Repositories, CI/CD pipelines, container images | Open source dependencies, including inside container layers |
| ReversingLabs Spectra Assure | Malware, tampering, suspicious behavior, exposed secrets | Finished software packages before release or deployment | Open source and commercial software, checked in compiled form |
Aikido Security, founded in 2022, has bases in Ghent, Belgium and San Francisco. It protects your code, cloud and running apps from one central system, which also covers autonomous pentesting (attack tests run by AI). And its big goal is self-securing software, meaning security that keeps testing and improving itself as your software is built and released.
Free install-time protection: Aikido Safe Chain sits in front of your package manager. At install, it stops malware, typosquats (fakes named one typo away from popular packages), compromised releases and dependency confusion (an attacker's public package that shares a name with your private one).
Wide tool support: It covers 16 package managers, including npm, yarn, pnpm, bun, pip, uv and poetry.
Safer new releases: Brand-new package versions are held for a waiting period you can configure, so they get a malware check first.
Repo malware checks: From there, once your repos are connected, Aikido checks every installed package for malware, switched on by default on Pro plans and above. On the Pro plan, new container images get scanned too.
AI tool coverage: On top of that, Device Protection extends install blocking to every developer device, covering package registries (the online stores packages come from), extensions and AI tools such as Claude Code.
If your team lives on GitHub, Socket is worth knowing. It guards your code against vulnerable and malicious dependencies.
Behavior-based detection: It reads what package code does and flags things like network calls, file access, shell commands and scrambled code meant to hide its purpose.
Install-time checks: Socket Firewall, a light filter between you and the package source, checks each package at install, on laptops, in CI (automated builds) and across your network.
Review-time alerts: Socket for GitHub warns about risky dependencies and malicious updates in pull requests. Socket says it flags malicious packages within minutes of publication.
Endor Labs keeps an eye on freshly uploaded open source packages and checks them for malicious code and risky behavior.
New upload scanning: It scans every new upload to npm, PyPI and other registries for typosquats, dependency confusion, install-time scripts (code that runs by itself the moment a package installs) and login stealers.
Shared rules: Package Firewall keeps risky packages off developer machines and CI pipelines, and AI coding agents like Claude Code and Cursor follow the same rules.
Flexible policies: Policies can also stop packages that aren't confirmed malware, using cooldown periods (a wait before new versions are allowed), license type or maintainer reputation.
Well, Sonatype Firewall works at the point of download. It blocks malicious open source components the moment you download them, across repositories, the network edge and endpoints (the computers people work on).
Hands-off quarantine: Suspicious components go into quarantine (held aside until confirmed safe) and are released automatically once they check out.
Network tool support: On top of that, it works with network tools like Zscaler to stop malware at the network edge.
Setup options: Sonatype Firewall Pro covers npm, Maven, PyPI and NuGet on repositories other than Nexus Repository, and Sonatype Repository Firewall is built for Nexus.
If your team runs on Artifactory, JFrog Curation checks packages, AI assets and IDE extensions (add-ons for code editors) as they're requested. Therefore, risky ones get stopped before they reach your developers.
Early threat research: JFrog's research team spots threats hidden through typosquatting and maintainer hijacking (an attacker taking over a real maintainer's account), often ahead of public databases.
New release hold: New versions sit out a 14-day immaturity hold, so the community has time to report a poisoned release.
Blocking at request: According to JFrog's Curation page, 99% of malicious components get stopped at the point of request.
Checkmarx Malicious Package Protection finds open source packages with known malware or suspicious behavior. Plus, it's part of the Checkmarx One platform.
Automatic responses: When it finds a malicious package, your policies can send alerts, block pull requests or break builds.
Broad ecosystem watch: It watches ecosystems including npm, PyPI, Maven, RubyGems, NuGet, Go, Rust and Swift for harmful packages, lookalike names, dependency confusion and hijacked maintainer accounts.
Human-reviewed findings: Its research team reviews flagged packages by hand before they go into the database. Checkmarx claims a proprietary database of 550,000+ malicious packages.
Mend SCA flags malicious open source packages right next to known vulnerabilities. Thus, your team sees both kinds of risk in one place.
Coverage beyond CVEs: Beyond known CVEs, it flags typosquats, dependency confusion attacks and packages carrying data-stealing or backdoor code.
Harmful code blocked: It blocks protestware (packages a maintainer changed to push a message or cause damage), data stealers and crypto miners.
Container coverage: In addition to your dependencies, it reaches into container images, so malicious packages packed inside them stay visible.
Finally, ReversingLabs looks at software from the other end. Spectra Assure breaks your software packages down to their compiled files to find malware, tampering and exposed secrets before release.
Release protection: It spots code tampering before a package leaves the build, and can stop a release that changes in a suspicious way.
Clear reporting: The SAFE report gives you an SBOM (an ingredient list of everything inside the software) plus a risk check.
Deep threat data: As per ReversingLabs, its threat intelligence database covers 400 billion files, with 16 proprietary malware detection engines.
Not sure which one fits your team? Well, each tool guards a different spot, so start with where you need cover most. And a tool missing from a bullet may still handle that job.
At install: Aikido Safe Chain and Device Protection, Socket Firewall and Endor Labs Package Firewall check packages before they land on your machines.
Where packages come in: JFrog Curation checks each request inside Artifactory, and Sonatype Firewall quarantines suspicious components as you download them.
In repos and builds: Aikido scans repos and containers, Socket warns in pull requests, Checkmarx can block pull requests or builds, and Mend SCA covers pipelines.
Brand-new releases: Aikido's waiting period, JFrog's immaturity hold, Endor Labs' cooldowns and Sonatype's quarantine slow fresh versions down, so your team gets time to check them.
Finished software: ReversingLabs Spectra Assure checks the built package itself, even if you don't have the source code.
Well, a bad package won't tell you it's bad, and a clean CVE scan won't spot it either. So the safest move is to check packages when they come in, while they sit in your code and before your software goes out.
Not sure where to start? If you want something free on day one that can grow with you into repos and containers, Aikido is a good place to begin.
AI tools to find & convert leads. 
Send emails at scale
