Supply Chain Attacks: How Trusted Software Turns Hostile
How attackers reach you through software and suppliers you already trust, what an SBOM really gives you, and which parts you can actually control.

In March 2024 a developer investigating why a login process was running slightly slower than expected traced the delay into a compression library. What he found, tracked as CVE-2024-3094, was malicious code present in the upstream source archives of xz starting with version 5.6.0 and affecting releases 5.6.0 and 5.6.1. The mechanism was deliberately obscured: the build process extracted a prebuilt object file from a disguised test file and used it to modify functions in the resulting library, so that any software linking against that library could have its data interactions intercepted and modified. The CVE record carries a CVSS version 3.1 base score of 10.0.
The detail that matters is not the technique. It is that the backdoor arrived through the normal, trusted path for getting software: an official release archive from a long-running project, on its way into mainstream distributions. Nothing was broken into. The channel worked exactly as designed, and it was carrying something it should not have been.
That is the whole shape of supply chain security. The attacker does not come through your perimeter; they come through something you already decided to trust.
The four places it goes wrong
Lumping these together is why supply chain security feels unmanageable. They are different problems with different defenses.
1. Dependency compromise
Modern applications are mostly other people's code. A typical web application pulls in hundreds of packages directly and indirectly, each maintained by someone you have never spoken to, updated automatically, and executed with your application's privileges.
Attackers reach that code in several ways. They take over a maintainer's account, often through credential reuse or a phished password, and publish a malicious version of a package people already depend on. They acquire an abandoned but widely used package from a maintainer who is relieved to hand it over. They publish packages with names close to popular ones, relying on typing mistakes. Or they exploit dependency confusion, where a build configured to check both a private registry and a public one can be tricked into preferring an attacker's public package that shares a name with an internal one.
The xz case is the most sophisticated documented example of the maintainer route, where the eventual contributor built up legitimate standing in the project over a long period before the malicious change appeared.
2. Build and distribution compromise
If an attacker controls the system that compiles and signs your software, the source code can be clean and the released artifact can still be malicious. Reviewing the repository will not show it, and the signature will be valid because it was produced by the real signing key.
The same applies to the update channel. Software that updates itself is a mechanism for executing remote code on every installation, protected mainly by code signing and the security of the infrastructure holding the keys.
This is the category the SLSA framework addresses. Its security levels describe progressively stronger guarantees about how an artifact was produced: that there is provenance describing the build, that the build ran on a hosted service rather than a developer's laptop, and that the build platform resists tampering by someone who can submit code. It is one of the few places in this field with a concrete, incremental specification to follow.
3. Service provider and vendor access
This is the most common form in practice and gets the least attention. Managed service providers, IT support contractors, monitoring vendors, backup services and software platforms hold credentials and remote access into their customers' environments. Compromising one provider can give an attacker a route into every customer, which makes them an efficient target.
No amount of code analysis helps here, because no malicious code is involved. The attacker uses a legitimate remote access tool with legitimate credentials. The controls that help are the unglamorous ones: knowing exactly which third parties hold access, scoping that access to what they genuinely need, requiring multi-factor authentication on their accounts into your systems, and logging their sessions where you can review them. Related to this, vendor email compromise is how a supplier's breach most often reaches a customer's finance team.
4. Hardware and firmware
Hardware implants during manufacture are the version of this that attracts headlines and the one almost no organization will encounter. The realistic hardware concerns are firmware that ships with known vulnerabilities and never gets updated, counterfeit components with unknown provenance entering through grey-market resellers, and devices arriving pre-configured with default credentials. Firmware updates for network equipment, management controllers and peripherals are frequently the oldest unpatched software in an organization, which makes them a better use of attention than implant scenarios.
What an SBOM gives you, and what it does not
The Log4j response in December 2021 is the clearest argument for software bills of materials. CVE-2021-44228 affected Apache Log4j 2 from version 2.0-beta9 through 2.15.0, excluding the security releases 2.12.2, 2.12.3 and 2.3.1, with the behavior fully removed from 2.16.0 onward. It carried a CVSS version 3.1 base score of 10.0, and CISA published dedicated guidance because of how widely the library was embedded.
The problem for most organizations was not patching. It was that they could not determine whether they were affected. Log4j was bundled inside commercial products, inside appliances, inside containers built from base images nobody had inspected. Weeks went into answering a question an inventory would have answered in an afternoon.
An SBOM is that inventory: a machine-readable list of components, versions and relationships for a piece of software. CISA coordinates work on formats and practices. Used well, it converts the next Log4j from a research exercise into a query.
Three honest limitations are worth holding onto. An SBOM describes composition, not security, so it tells you a component is present and nothing about whether it is reachable or exploitable in your configuration. Its accuracy depends entirely on how it was generated, and one produced by scanning a built artifact often differs from one derived from a dependency manifest. And an SBOM you receive and file is worthless; the value exists only if you can search across all of them when an advisory lands.
For organizations buying software rather than writing it, the practical move is to ask for an SBOM during procurement and to have somewhere to put it. US federal software suppliers now attest to secure development practices drawn from NIST's Secure Software Development Framework, which makes "how was this built" a reasonable procurement question rather than an unusual one.
What you can realistically control
NIST's SP 800-161 Rev. 1 is the comprehensive treatment of cybersecurity supply chain risk management, and the NCSC's supply chain guidance is the more approachable version. Both are worth reading, and both are aimed at organizations with more capacity than most have. The shorter list below is what tends to be achievable.
If you write software:
- Pin versions and commit lock files, so a build is reproducible and an upstream publish cannot silently change what you ship.
- Run dependency scanning in the build, and fail on known-vulnerable versions rather than generating a report nobody reads.
- Review what new dependencies cost you. A package with one maintainer, no recent releases and forty transitive dependencies is a decision, not a convenience.
- Treat the build system as production. It holds signing keys and deployment credentials. Restrict who can modify pipeline configuration, and require review for changes to it.
- Remove long-lived credentials from pipelines in favor of short-lived, scoped tokens issued at run time.
- Resolve internal packages from internal registries only, which closes dependency confusion.
- Generate an SBOM at build time and keep it somewhere searchable.
If you buy software and services:
- Keep a list of who has access to what. Every supplier with remote access, API credentials or an account in your systems, with an owner and a review date.
- Scope that access down. A monitoring vendor rarely needs administrative rights, and a support contractor rarely needs standing access rather than access on request.
- Require strong authentication for supplier accounts, and insist their access runs through your identity provider where possible so you can revoke it.
- Put incident notification in the contract, with a defined window, and know who at the supplier you would call.
- Ask about SBOMs and development practices during procurement, when you have leverage, rather than after signing.
- Be able to patch fast. This is the control that pays out in every scenario above, and it is covered in our vulnerability management guide.
Detection deserves a line of its own. Supply chain compromise defeats prevention by design, so the realistic goal is noticing the behavior that follows: a build server making outbound connections it has never made, a signed application reading credential stores, a vendor account active at an unusual hour from an unfamiliar location. Our endpoint security guide covers the tooling and threat intelligence covers choosing what behaviors to look for.
A supplier's compromised mailbox is the most common version of this. Changed bank details arriving in a real thread from a real address is the classic form. Paste the message into CyberWatch AI Scan for a free check, and verify payment changes by phone on a number you already held.
Keeping it in proportion
Supply chain attacks are genuinely hard to prevent and they are not how most organizations get compromised. Phishing, unpatched internet-facing systems and stolen credentials remain far more common, and effort spent on them buys more safety per hour.
What supply chain thinking adds is a correction to an assumption. The code and the people you already trust are part of your attack surface, and most organizations have never enumerated either. Start with the list of third parties holding access to your systems, because it is usually shorter than expected, almost always more permissive than anyone intended, and the one part of this problem you can fix without anyone else's cooperation. For flaws discovered with no fix yet available, see our guide to zero-day vulnerabilities.
Frequently asked questions
What is a software supply chain attack?
An attack that reaches its victims through software or a supplier they already trust, rather than by breaking in directly. The attacker compromises a dependency, a build system, an update channel or a service provider, and the malicious code arrives through a legitimate installation or update.
Why are these attacks effective?
Because the delivery mechanism is trusted by design. Signed updates from a known vendor, dependencies pulled automatically during a build, and remote access held by a managed service provider all bypass the controls aimed at outsiders. One compromise can also reach thousands of downstream organizations.
What is an SBOM?
A software bill of materials: a machine-readable list of the components in a piece of software, with versions and relationships. Its main value is answering the question that paralyzed organizations during the Log4j response, which was simply whether they used the affected component at all.
Does an SBOM make software secure?
No. It is an inventory, not an assessment. It tells you what is in the box, which is the prerequisite for knowing whether a newly disclosed vulnerability affects you. It says nothing about whether the components are configured safely or whether the code is well written.
Can a small organization do anything about supply chain risk?
Yes, though mostly in the second half of the chain rather than the first. You cannot audit an open-source maintainer, but you can control which suppliers hold remote access to your systems, limit what their accounts can reach, pin dependency versions, and be able to patch fast when something is disclosed.
What is a dependency confusion attack?
A technique where an attacker publishes a package to a public registry using the name of a company's internal private package. If the build system is configured to check the public registry as well, it may pull the attacker's version instead of the internal one.
Are hardware supply chain attacks real?
Documented cases exist, mostly involving firmware and counterfeit components rather than the implanted chips that generate the most headlines. For almost every organization, firmware updates, device provenance and secure boot are the practical concerns rather than tampering during manufacture.
How do I assess a supplier's security?
Ask what access they need and reduce it, rather than relying only on a questionnaire. Contracts should require notification of incidents affecting your data within a defined period, and the access you grant should be scoped, time-limited and logged where you can see it.
What single change helps most?
Knowing what you run, in enough detail to answer a disclosure question within hours. Organizations that handled the Log4j response well were almost always the ones who already had a dependency inventory, not the ones who worked hardest that week.
Sources
- SP 800-161 Rev. 1, Cybersecurity Supply Chain Risk Management Practices for Systems and Organizations, US National Institute of Standards and Technology
- SP 800-218, Secure Software Development Framework (SSDF) Version 1.1, US National Institute of Standards and Technology
- Software Bill of Materials (SBOM), US Cybersecurity and Infrastructure Security Agency
- SLSA Security levels, Supply-chain Levels for Software Artifacts
- Supply chain security guidance, UK National Cyber Security Centre
- Reported Supply Chain Compromise Affecting XZ Utils Data Compression Library, CVE-2024-3094, US Cybersecurity and Infrastructure Security Agency
- CVE-2024-3094, CVE Program
- CVE-2021-44228, CVE Program
- Apache Log4j Vulnerability Guidance, US Cybersecurity and Infrastructure Security Agency
- Secure Software Development Attestation Form, US Cybersecurity and Infrastructure Security Agency


