← all systems · see it in The System
Supply chain security
Activities that keep trust in the parts a system is built from and in the way they travel: for goods, the security of the transport and logistics chain (credentialing participants, screening cargo, seals, inspection on entry) against terrorism, piracy and theft; for computing, the integrity of the components, libraries, tools, build systems, hardware and vendors that make up a product, against malicious code or circuits injected upstream to reach users downstream (src:wp:supply-chain-security; src:ohm-2020-backstabber). Walked as the practice with the people, organisations, standards and laws that carry it.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | 1974 (Karger and Schell, trusting trust); 2001 (C-TPAT, after 9/11) |
| born precision | year |
| boundary | Inside: the activities that keep parts and their journeys trustworthy: credentialing suppliers and maintainers, inventories of components (bills of materials, SBOMs), signing and provenance (code signing, in-toto layouts, SLSA provenance), reproducible and diverse builds, isolated build systems, screening and testing of incoming parts (package scanning, chip testing, cargo screening), seals and tracking in transit, inspection on entry, and the response when a part turns out to be compromised (revocation, rebuilding, takedown); plus the standards, programmes and laws that carry them (C-TPAT, ISO 28000, O-TTPS, SSDF, SLSA, EO 14028). Outside: the supply chains themselves (goods, software, hardware), the attackers, the vulnerabilities, the products built, and the organisations' wider security. |
| nested level | A practice carried by organisations, maintainers, registries and states (rung 12), working on supply chains; the part of information security and of supply-chain management that deals with what comes in from others. Overlaps secure software development, whose SSDF covers protecting the software and its build. |
| scale | technological |
Measures (29) published science
| characteristic clock: time a malicious package is available before public report | 209 mean (median 67, max 1,216) days |
| SolarWinds trojanized update dormancy | up to 14 days |
| XZ campaign duration | over two years (2021 to 2024) |
| Log4Shell time unnoticed | 2013 to November 2021 (about 8) years |
| malicious packages in the Backstabber dataset (npm, PyPI, RubyGems, 2015-2019) | 174 packages |
| share of malicious packages using typosquatting | 61 % |
| average Levenshtein distance of a typosquat to its target | 2 edits |
| event-stream dependents at takeover | 1,600 packages; 1.5 million weekly downloads |
| CCleaner trojanized downloads before detection | about 2.3 million (over a month) downloads |
| implicit trust per average npm package | 79 packages, 39 maintainers count |
| maintainers whose accounts reach more than half of npm | 20 maintainers |
| share of npm packages depending on known-vulnerable code | up to 40 % |
| attack vectors / incidents / safeguards in the OSS supply-chain taxonomy | 107 / 94 / 33 count |
| Debian packages reproducible | over 95% of 30,000+ (after seven years) |
| reproducible builds: experts rating utility very high / cost high | 58.8 / 70.6 % |
| commercial codebases with open-source code; share of code open source | 96 / 76 (1,703 codebases, 2023) % |
| PyPI projects listing a signature | about 4 % |
| SolarWinds entities affected / US agencies confirmed compromised | about 18,000 / fewer than 10 count |
| historical supply chain attacks addressable by in-toto deployments | 83-100 % |
| XZ commits carrying the attack | 8 commits |
| Log4Shell enterprise cloud environments affected | 93 % |
| C-TPAT members (2014) | 10,854 (7 at launch, November 2001) members |
| share of US import value carried by C-TPAT importers | about 54 (4,315 importers) % |
| Target fourth-quarter 2013 profit fall after the HVAC-vendor breach | 46 % |
| reported year-over-year rise in software supply chain attacks (vendor report) | 650 (2021), 430 (2020) % |
| reported rise in supply chain attacks, 2018 (vendor report) | 78 % |
| substandard and falsified medicines prevalence (LMICs) | 13.6 (95% CI 11.0-16.3; 67,839 samples) |
| mitigation tasks missing from all ten frameworks | 3 of 73 tasks mapped |
| per-hand-off yield of a security fix down the dependency chain | below 0.5 (more than half of dependents still affected after the fix) fraction |
Open contentions (4) open contention
Where sources or readings disagree. Both stay on the map.
- C-1550 Who was behind the SolarWinds and XZ attacks?
booked: The map keeps attribution open where the sources disagree.
the other reading: XZ: any state or a resourced non-state actor (Claburn); SolarWinds: China suggested by President Trump - C-1551 How fast are software supply-chain attacks rising?
booked: The map keeps the vendor-reported rises as reported, and notes they are vendor figures.
the other reading: counts depend on each vendor's definition and collection; the term has no agreed definition - C-1552 When did supply-chain security begin?
booked: The map keeps both starts: the 1974 'trusting trust' warning for computing, and 2001 for cargo.
the other reading: born with 9/11 as a structured practice; before that the concern of insurers and risk managers; born with trade itself (no source read) - C-1553 Is cargo security the same system as software supply-chain security?
booked: The map books cargo as a variant for now, and keeps the reading that they are two systems.
the other reading: two systems: cargo security and ICT supply-chain security, sharing only a name; the term has no agreed definition ('a broad term without a universally agreed upon definition')
Partners and parts published science
Built from / builds (12)
- Code signing
- Reproducible builds
- Diverse double-compiling
- in-toto layouts and link metadata
- Supply-chain Levels for Software Artifacts (SLSA)
- Hardware Trojan detection (logic testing, side-channel analysis)
- Cargo and logistics security
- Customs-Trade Partnership Against Terrorism
- ISO 28000 security management systems
- Package registries and their policies
- Software bill of materials
- Attackers vs defenders over the trust consumers place in packages
Sits inside / holds (3)
Trades with (valence) (14)
- Open-source software via sys:software-supply-chain · Supply chain security gain
- Incident response via sys:software · Supply chain security mixed
- Threat actor via sys:software-supply-chain · Supply chain security loss
- Security policy and governance via sys:supply-chain-security · Supply chain security mixed
- Human (Homo sapiens) via sys:human · Supply chain security gain
- Law (rules and law) via sys:law · Supply chain security gain
- Supply chain (production network) via sys:supply-chain · Supply chain security gain
- Economy via sys:economy · Supply chain security gain
- Digital signatures and certificates via sys:digital-signature · Supply chain security gain
- Organization via sys:organization · Supply chain security gain
- Computer hardware via sys:computer-hardware · Supply chain security gain
- Software vulnerability via sys:software-vulnerability · Supply chain security loss
- Secure software development via sys:software · Supply chain security gain
- Software supply-chain security via sys:software-supply-chain · Supply chain security gain
expressed by (18)
- Templating / Standardization
- Sensing/Scanning
- Mapping/Modeling
- Uncertainty Reduction/Learning
- Trail-Marking / Externalized Memory (Stigmergy)
- Social Learning / Public Information Use
- Reactive Response
- Proactive Defense
- Physical/Virtual Barrier
- Technique-Based Identification (a.k.a. Technique-Based Detection)
- Self-Monitoring
- Validation of Identity/Intention
- Validation of Effectiveness
- Enclave/Segregation
- Resilience & Repair
- Redundancy & Diversity
- Triage & Prioritized Allocation
- Quality Control & Error Correction
floor filled by (23)
- Law (rules and law)
- Supply chain (production network)
- Economy
- Open-source software
- Threat actor
- Information security
- Organization
- Computer hardware
- Software vulnerability
- Code signing
- Secure software development
- Reproducible builds
- Diverse double-compiling
- in-toto layouts and link metadata
- Supply-chain Levels for Software Artifacts (SLSA)
- Hardware Trojan detection (logic testing, side-channel analysis)
- Cargo and logistics security
- Customs-Trade Partnership Against Terrorism
- ISO 28000 security management systems
- Package registries and their policies
- Software bill of materials
- Software supply-chain security
- Package manager
represents (1)
Kinds (4)
- Cargo and logistics security
- Software supply-chain security
- Hardware supply-chain security
- Medicine supply-chain security (substandard and falsified medicines)
Notes published science
- XZ Utils, a compression library inside most Linux systems, was backdoored by someone who spent over two years earning a lone volunteer maintainer's trust; it was caught because one developer noticed SSH logins using too much CPU and chased it (src:wp:xz-utils-backdoor, src:przymus-2025-xz).
- Installing one average npm package means trusting 79 other packages and 39 strangers, and 20 maintainer accounts could reach more than half of the whole ecosystem (src:zimmermann-2019-npm).
- Bacteria fall for the same trick: a nutrient importer that the cell builds itself can be fed a 'Trojan horse' and end up importing its own inhibitor, much as a trusted update channel delivered SolarWinds' signed backdoor (src:peukert-2023-tonb, src:hamer-2025-closing-chain).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
supply chain security as practitioners and researchers describe it: the parts a system is built from and where they came from, up to products whose parts can be trusted (reader's view)
- L1: The parts and their sources: packages and libraries, build tools and build systems, chips and IP cores, vendors with access, cargo in transit, and the people who maintain them
- L2: Intake and checking: dependency managers, signature verification, rebuilds, chip tests, credentialing, seals and screening
- L3: Findings: typosquats, mismatched binaries, invalid or misused signatures, Trojans, counterfeit parts, vulnerable dependencies
- L4: The inventory and provenance picture: SBOMs, layouts and link metadata, provenance
- L5: Trust decisions: pin, verify, accept or reject; consensus of independent builders
- L6: Organisation and governance: frameworks, programmes, standards, registry policy and law
- L7: Purpose and response: products verifiable from inception to deployment, legitimate trade facilitated; revocation and rebuilding when trust fails
the boardroom (shared floors)
- L4: The inventory and provenance picture: what is in the product and where each part came from (SBOM, layout and link metadata, provenance, the implicitly trusted code base)
- L5: Trust decisions: accept, pin, verify, reject; reproducible-build consensus; diverse double-compiling; MFA and isolated hardened builds
- L6: The organisation's supply-chain risk management: frameworks and programmes (SSDF, SLSA, ISO 28000, C-TPAT, O-TTPS), stakeholder roles, registry policy; mandates join here
- L7: Purpose: products whose parts can be trusted from inception to deployment, and legitimate trade facilitated; response when that trust fails
way in: Anomalies in running systems
- L1: The installed product running: a trojanized update dormant for up to two weeks; a backdoored library loaded by sshd
- L2: Noticed: unexpected CPU use and Valgrind errors in SSH logins (XZ); system monitoring
- L3: Investigated and reported (Freund to Debian and Openwall; FireEye's discovery of SolarWinds)
way in: Goods in transit (cargo and containers)
- L1: Cargo moving through shipper, carrier and consignee; theft and shrinkage possible anywhere
- L2: Seals and locks in transit; tracking (RFID, GPS); advance notification of contents to the destination
- L3: Screening and inspection: containers screened at foreign ports (CSI), cargo inspected on entry
way in: Hardware parts (chips, IP cores, devices)
- L1: Chips and components from outsourced fabrication, bought IP cores and brokers; possible Trojans or counterfeits
- L2: Testing: logic testing to activate rare events; side-channel measurement (power signature) without triggering
- L3: Deviation found or missed: small Trojans hidden by process variation; triggers on rare conditions
way in: Law, mandates and markets
- L1: Mandates: EO 14028 (12 May 2021); EU Community Customs Code (2008); registry signing policies
way in: People in the chain (maintainers and contributors)
- L1: Maintainers and contributors, partly unknown or anonymous; a few accounts reach most of npm
- L2: Contributions, reviews and requests (event-stream handed over on request; XZ's years of patches and reviews)
- L3: Trust granted: maintainer permissions, publishing rights
way in: Signatures and provenance records
- L1: Signed artifacts arriving: code signatures, PGP signatures in registries, in-toto link metadata, SLSA provenance
- L2: Verification: signature checked with the author's key or certificate; checksum confirms no alteration since signing
- L3: Checked against the expected: link metadata against the layout; a valid signature can still sit on a trojanized product (SolarWinds)
way in: The build (source to binary)
- L1: The build chain: version control, build farm, CI/CD, packaging; compromising any single step can modify the software
- L2: Independent rebuilds from the same source (reproducible builds) or with a second, trusted compiler (diverse double-compiling)
- L3: Mismatch or match: binaries that do not match their source; build provenance (metadata on how outputs were generated)
way in: Third-party components (packages and libraries)
- L1: Packages published in registries (npm, PyPI, RubyGems, Maven); 96% of commercial codebases use open source, 76% of their code
- L2: The dependency manager resolves, downloads and installs packages (and runs install scripts) automatically
- L3: The transitive tree: an average npm package brings implicit trust in 79 packages and 39 maintainers; a typosquat sits two edits from its target
way in: Vendors and suppliers (organisations with access)
- L1: Suppliers and their access: adversaries infiltrate trusted suppliers and vendors (Target via an HVAC contractor's credentials)
- L2: Credentialing participants; certification programmes (C-TPAT, Common Criteria EAL 4, ISO 28000)
- L3: Risk standing: certified firms considered low risk
way in: Vulnerability and threat reports
- L1: Disclosures about components: Log4Shell (unnoticed since 2013, disclosed 2021), CTI reports on SolarWinds, Log4j, XZ
- L2: Matched against the inventory: with an accurate SBOM you immediately know whether you are impacted
- L3: Impact judged: SBOM-based management rarely considers actual exploitability
way in: What practitioners know (inner)
- L1: Case histories and principles: Thompson's lesson; Saltzer and Schroeder's principles read against SolarWinds and Log4j
- L2: Taught and surveyed: expert and developer surveys (17 experts, 134 developers)
- L3: Organised as an attack tree: 107 vectors, 94 incidents, 33 safeguards
way out: Block, pin and remove
- L1: The registry and the dependency tree changed
- L3: Malicious package reported and removed (on average 209 days after publication)
- L5: Decision: pin versions, disable install scripts, reject a package that fails verification
way out: Respond: revoke, rebuild, decontaminate
- L1: Systems taken offline and decontaminated over months; a safe version restored
- L2: Code-signing certificate revoked (SolarWinds); distributions rebuild their packages (XZ)
- L7: Response chosen after a compromise
way out: Seal, release or inspect cargo
- L1: Cargo released, sealed or held
- L3: Expedited processing and fewer customs examinations, or inspection on entry
- L6: Certified low-risk status (C-TPAT) or a risk flag
way out: Sign, attest and publish inventories
- L1: Signed artifacts and SBOMs published for the next link downstream
- L3: Provenance generated by the CI/CD pipeline; SBOMs generated in machine- and human-readable formats
- L6: Policy: registries mandate signing; standards require SBOMs and provenance