Attack clock vs. compliance clocks (log scale)
Bars end where each clock runs out. Deadlines start at different events: awareness, classification, materiality, or patch release.
Software compliance is meeting, and proving you meet, the security and legal controls that laws, standards, and contracts require of the software you ship.
Also known as security compliance, regulatory compliance, cybersecurity compliance, IT compliance, software supply chain compliance
Software compliance, also called security or regulatory compliance, is meeting, and proving you meet, the requirements that laws, industry standards, and contracts place on the software you ship, open source dependencies included. Frameworks such as SOC 2, ISO 27001, PCI DSS, and the EU Cyber Resilience Act define the controls, and audits test evidence that they operate.
The mechanism is a chain. A framework states a requirement, the organization maps it to a control, the control produces evidence, and an assessor samples that evidence. For third-party code, almost every chain is built around known vulnerabilities. The PCI Security Standards Council states the objective of its component inventory requirement plainly: "Known vulnerabilities in third-party software components cannot be exploited in bespoke and custom software." A malicious package is not a known vulnerability. When keyv@6.0.0 reached npm with a credential-stealing install hook on August 4, 2026, during the keyv and cacheable compromise, it carried no CVE and valid build provenance, and it ran before any patch clock could start.
Our position: compliance measures whether a process exists, not what a dependency does when it installs. A clean SOC 2 report or an ISO 27001 certificate is good evidence of discipline and weak evidence of supply chain safety. Treat compliance frameworks as the floor and the budget line. Then put a control in front of installs that inspects package contents. We are engineers, not lawyers, and nothing here is legal advice.
Software compliance is the part of organizational compliance that governs code: how it is written, changed, inventoried, patched, and shipped. It sits inside governance, risk, and compliance (GRC), the discipline that ties a company's controls to its risk register and its legal obligations. Software compliance is where GRC meets the build system.
Every compliance regime has the same four parts:
The assessment type decides what a result means. A SOC 2 report is an attestation: a licensed CPA firm gives an opinion on controls measured against the AICPA's Trust Services Criteria. A Type 1 report covers control design at a point in time, and a Type 2 report covers operating effectiveness over a period. SOC 1 covers controls relevant to financial reporting, and SOC 3 is a short general-use version of SOC 2. None of them is a certification.
ISO/IEC 27001 is a certification. An accredited certification body audits an information security management system (ISMS), the documented set of policies, risk assessments, and controls an organization runs. The 2022 revision cut Annex A from 114 controls to 93, grouped into four themes, and certificates issued against the 2013 version expired on October 31, 2025 (Kiwa). The Statement of Applicability (SoA) is the ISMS document that lists which Annex A controls apply and why.
PCI DSS is a contractual standard. Card brands impose it through acquiring banks on anyone who stores, processes, or transmits payment card data. The acquirer or payment brand decides how each entity validates: a Report on Compliance from a Qualified Security Assessor, or a Self-Assessment Questionnaire (SAQ) (PCI SSC). Laws such as GDPR, NIS2, DORA, and the Cyber Resilience Act are enforced by regulators instead, with fines. US federal systems go through assessment and authorization (A&A), which ends in an authorization to operate.
Compliance auditing is the assessment step, and the whole cycle is expensive. In a summary of responses to the White House's 2023 request for information on regulatory harmonization, the Financial Services Sector Coordinating Council reported that many sector CISOs spend 30 to upwards of 50 percent of their time on regulatory compliance.
Compliance and security overlap, but they answer different questions. Compliance asks whether a required control existed and operated during the audit period. Security asks whether an attacker can get in today.
Compliance frameworks handle open source dependencies as part of the organization's own software, not as a supplier relationship. Supplier clauses assume a contract. NIS2 Article 21(2)(d) asks for "supply chain security, including security-related aspects concerning the relationships between each entity and its direct suppliers" (NIS2, Article 21). The maintainer of a transitive npm dependency is nobody's direct supplier and signs nothing. That is why third-party risk management questionnaires reach vendors and almost never reach the packages in a lockfile.
Dependency obligations arrive instead through vulnerability management, change management, and inventory. This table shows where they live in the frameworks most teams face:
| Framework | Clause | What it asks for | Evidence auditors usually accept |
|---|---|---|---|
| SOC 2 | CC7.1 | Detect "susceptibilities to newly discovered vulnerabilities"; a point of focus adds detecting "unknown or unauthorized components" | Periodic vulnerability scans, remediation tickets |
| SOC 2 | CC8.1, CC9.2 | Authorize, test, and approve changes; assess vendor risk | Pull request approvals, CI logs, vendor reviews |
| ISO/IEC 27001:2022 | A.8.8, A.5.21, A.8.28 | Management of technical vulnerabilities; information security in the ICT supply chain; secure coding | Vulnerability procedure, scan output, supplier register |
| PCI DSS v4.0.1 | 6.3.2, 6.3.3 | Inventory of bespoke software and its third-party components; critical and high patches within one month | Component inventory or SBOM, patch records |
| NIS2 | Art. 21(2)(d), (e) | Supply chain security with direct suppliers; vulnerability handling and disclosure | Supplier risk assessments, disclosure policy |
| CMMC Level 2 | NIST SP 800-171 3.4.1 | Baseline configurations and inventories of systems, including software | System and software inventory |
| EU Cyber Resilience Act | Art. 13(5), Annex I | Due diligence on third-party components, including free and open source; an SBOM | SBOM, component checks |
| FDA FD&C Act | §524B(b)(3) | SBOM for cyber devices, including open-source components | SBOM in the premarket submission |
Read down the evidence column and a pattern appears: a list of components, and a list of CVEs matched against it. SOC 2's CC7.1 comes from the AICPA's Trust Services Criteria, and its point of focus on "unknown or unauthorized components" is the closest any of these frameworks comes to naming a malicious package. Only two entries name open source explicitly: the FDA's SBOM requirement and the Cyber Resilience Act. The CRA goes further and puts due diligence on the manufacturer that integrates each component.
As of November 2026, two legal regimes require a software bill of materials by name: section 524B of the US Federal Food, Drug, and Cosmetic Act for cyber devices, and the EU Cyber Resilience Act. PCI DSS requires one in substance without using the term. Everywhere else, an SBOM is optional or absent.
| Regime | SBOM status | What it must cover | In force |
|---|---|---|---|
| FDA, FD&C Act §524B(b)(3) | Required by name | "commercial, open-source, and off-the-shelf software components" of a cyber device | Premarket submissions since March 29, 2023 |
| EU Cyber Resilience Act, Annex I Part II(1) | Required by name | "at the very least the top-level dependencies" in a machine-readable format | Applies from December 11, 2027 |
| PCI DSS v4.0.1, Requirement 6.3.2 | Required in substance; no format named | Bespoke and custom software and the third-party components inside it | Mandatory since March 31, 2025 |
| US federal procurement, OMB M-26-05 | Optional, at agency discretion | Whatever an agency's risk assessment calls for | Since January 23, 2026 |
| SOC 2, ISO 27001, NIS2, DORA, GDPR, HIPAA, CMMC | Not required | Asset or system inventories, not software components | Not applicable |
The FDA text comes from 21 U.S.C. § 360n-2, which took effect 90 days after its December 29, 2022 enactment. The PCI DSS requirement became mandatory when its future-dated requirements took effect on March 31, 2025 (requirement text). The federal row changed in January 2026, when OMB rescinded the federal software attestation and SBOM mandates in M-22-18 and M-23-16. HIPAA may move: a Security Rule overhaul proposed on January 6, 2025 would add a technology asset inventory, but it was still not final as of June 2026 (North Privacy Advisors).
Guidance is moving faster than law. CISA's August 22, 2025 draft update to its SBOM minimum elements adds component hash, license, tool name, and generation context.
Verdict: if you sell cyber devices in the US or products with digital elements in the EU, an SBOM is a legal deliverable, and it should cover the full resolved graph rather than the Cyber Resilience Act's top-level floor. If you handle card payments, a CycloneDX or SPDX SBOM generated on every build is the cheapest way to answer Requirement 6.3.2. Everyone else should generate one anyway, because it is the first thing incident response needs. Do not expect it to prove much on its own.
Which SBOM you generate decides what it can prove. We built a project with Express as a dependency and ESLint as a dev dependency, then generated two CycloneDX SBOMs with npm 10.9.9 on November 8, 2026:
$ npm ls keyv --all
sbomdemo@1.0.0
`-- eslint@9.39.5
`-- file-entry-cache@8.0.0
`-- flat-cache@4.0.1
`-- keyv@4.5.4
$ npm sbom --sbom-format cyclonedx | jq '.components | length'
158
$ npm sbom --sbom-format cyclonedx --omit dev | jq '.components | length'
68
$ npm sbom --sbom-format cyclonedx --omit dev | jq '[.components[] | select(.name=="keyv")] | length'
0The runtime SBOM, the one you hand a customer or a regulator, drops 90 of 158 components, including the whole ESLint chain. That chain matters. Our analysis of the keyv compromise traced a common path to victims: eslint to file-entry-cache to flat-cache to keyv, and trojanized releases of flat-cache and file-entry-cache were published in the same campaign. This demo resolves the clean keyv@4.5.4. ESLint runs in CI, where publish tokens and cloud credentials live, yet a runtime SBOM never mentions it. A runtime SBOM is the right artifact for a customer and the wrong one for scoping a supply chain incident.
A clean audit misses a malicious package because the controls it tests look for known vulnerabilities and approved changes, and a fresh malicious release is neither. Here is keyv@6.0.0 measured against them.
The npm registry recorded keyv@6.0.0's publish time as 2026-08-04T09:35:00.763Z. The version's preinstall hook downloaded a standalone Bun runtime and ran an obfuscated second stage that harvested cloud, Vault, Kubernetes, GitHub Actions, and npm credentials. Socket's malware detection flagged it about six minutes after publication, and across the campaign Socket detected affected artifacts an average of five minutes and 18 seconds after publication. GitHub published a malware advisory the same day. This is what npm's own advisory endpoint returns for that version today:
$ curl -s -X POST https://registry.npmjs.org/-/npm/v1/security/advisories/bulk \
-H 'content-type: application/json' -d '{"keyv":["6.0.0"]}' | jq .
{
"keyv": [
{
"id": 1130747,
"url": "https://github.com/advisories/GHSA-3p9h-f68w-m6fx",
"title": "Malware in keyv",
"severity": "critical",
"vulnerable_versions": "=6.0.0",
"cwe": ["CWE-506"],
"cvss": { "score": 0, "vectorString": null }
}
]
}The record has no CVE, its weakness is CWE-506 (embedded malicious code), and its CVSS score is 0 with no vector. A scanner policy that reads the severity label sees critical. A policy that gates on CVSS score, such as "fail the build at 7.0 or above," sees zero. The advisory itself is blunt: "Any computer that has this package installed or running should be considered fully compromised."
Now run the usual controls against it:
| Control | Typical clause | What it checks | Result for keyv@6.0.0 |
|---|---|---|---|
| Component inventory or SBOM | PCI DSS 6.3.2, CRA Annex I | Is the package listed? | Listed only if the SBOM includes dev and CI dependencies. A listing is not a verdict. |
| Vulnerability scanning | SOC 2 CC7.1, ISO 27001 A.8.8 | Does the version match a known vulnerability? | No CVE at publish time. The later malware advisory has none either. |
| Patch window | PCI DSS 6.3.3 | Are critical patches installed within one month? | Irrelevant. The payload ran during install. |
| Change management | SOC 2 CC8.1 | Was the change reviewed and approved? | A lockfile bump shows a version and an integrity hash, not the new preinstall hook. |
| Build provenance | NIST SSDF PS.3, SLSA | Was it built by the expected workflow? | Passed. The legitimate release workflow built source the attacker had already modified. |
| Analysis of package contents | No framework names it | What does installing this version do? | Flagged about six minutes after publication. |
Package provenance is the row that surprises people. As our keyv analysis put it, "provenance attests build integrity, not source integrity." The NIST SSDF and SLSA ask for exactly that evidence, and it was valid. The vulnerability evidence is thinning too. In April 2026, NIST stopped enriching most CVEs in the NVD, keeping enrichment only for known exploited, federal, and EO 14028 critical software.
After an incident, the evidence trail decays. As of November 8, 2026, keyv's registry document still lists a publish time for 6.0.0 but no longer lists the version itself, so a fresh install cannot reproduce what ran. The axios compromise showed the same problem from the auditor's side: tools fetched with npx resolved the malicious version outside any lockfile, and as our analysis put it, "The absence of evidence after the fact is not strong evidence of absence."
Most compliance reporting deadlines start when an organization becomes aware of an incident. Detection latency therefore decides how much of a compliance clock is left, and how much damage happened before it started.
Attack clock vs. compliance clocks (log scale)
Bars end where each clock runs out. Deadlines start at different events: awareness, classification, materiality, or patch release.
The figure plots each deadline on a logarithmic scale from one minute to one month. The values, with sources:
| Event or deadline | Time | Clock starts at | Source |
|---|---|---|---|
| keyv@6.0.0 preinstall hook executes | During install | Install | Socket analysis |
| Socket flags keyv@6.0.0 | About 6 minutes | npm publish | Socket analysis |
| DORA initial notification of a major incident | 4 hours, and no later than 24 hours after awareness | Classification as major | DORA incident reporting standards |
| EU Cyber Resilience Act and NIS2 early warning | 24 hours | Awareness | CRA Art. 14; NIS2 Art. 23(4)(a) |
| GDPR breach notification; NIS2 and CRA notification; APRA CPS 234 | 72 hours | Awareness | GDPR Art. 33(1); NIS2 Art. 23(4)(b); CPS 234 |
| SEC Form 8-K Item 1.05 | 4 business days | Materiality determination | SEC cybersecurity disclosure rules |
| PCI DSS critical and high patches | 1 month | Patch release | PCI DSS 6.3.3 |
The sources: DORA's timelines are summarized by Morgan Lewis. The NIS2 deadlines are in Article 23, and GDPR's 72 hours are in Article 33. The Cyber Resilience Act's reporting obligations have applied since September 11, 2026. The SEC deadline has applied to incidents since December 18, 2023 (Hunton).
Two things follow. First, the patch window in PCI DSS 6.3.3 was designed for vulnerabilities that wait for an attacker. A malicious install hook does not wait. Second, awareness-based clocks seem to reward not looking, but they do not. The Cyber Resilience Act's due diligence duty and SOC 2's monitoring criteria both make "we never checked" a finding of its own.
Good compliance evidence is a by-product of controls that would have caught the attack. Each of these satisfies an auditor and also changes the outcome of an incident. Each has a cost.
Inventory the resolved graph on every build. Generate a CycloneDX or SPDX SBOM from the lockfile, including dev and CI dependencies, and keep one per release. This covers PCI DSS 6.3.2, the Cyber Resilience Act, and the FDA. It still misses tools run with npx or installed globally, which never touch the lockfile.
Triage CVEs by reachability and write the result down. Reachability analysis checks whether your code can call the vulnerable function, and a VEX document records the verdict in a form a buyer or auditor can read. Socket's CycloneDX export can carry reachability and VEX data in the SBOM itself when include_vulnerabilities is set; the endpoint is in beta (Socket docs). Static reachability is conservative. It cannot see code loaded through reflection or configuration it does not model, so treat a trace as evidence, not proof.
Gate installs on behavior, not just on CVEs. Inspect what each new version does before it runs: new install scripts, obfuscated code, network and credential access. That is the control that maps to SOC 2's "unknown or unauthorized components" and to the Cyber Resilience Act's due diligence. Socket Firewall intercepts package requests on their way to the registry and blocks versions already identified as malicious before they download or run, including during a fresh install or an npx execution. A block list only stops what has been identified, so the minutes between publish and detection remain open. Release-age cooldowns narrow that gap, at the cost of delaying legitimate fixes, and as we noted during the axios compromise, they are not standardized and can often be overridden in configuration.
Keep evidence you can replay. Retain lockfiles, build logs with resolved versions, and scan results tied to a commit. The registry will remove the malicious version, and your auditor will ask what ran.
Licenses are the other half of software compliance, with their own obligations and failure modes; we cover them under license compliance.
Most teams answer to several of these at once. The table lists what each one means for software and dependencies, dated as of November 2026:
| Framework | Applies to | Software and dependency angle | Status |
|---|---|---|---|
| SOC 2 (AICPA) | Service organizations, usually at customer request | CC7.1 vulnerability monitoring, CC8.1 change management, CC9.2 vendor risk | 2017 criteria with points of focus revised in 2022 |
| ISO/IEC 27001:2022 | Any organization certifying an ISMS | A.5.19 to A.5.23 supplier controls, A.8.8 vulnerabilities, A.8.28 secure coding | 2013 certificates expired October 31, 2025 |
| PCI DSS v4.0.1 | Anyone storing, processing, or transmitting cardholder data | 6.3.2 component inventory, 6.3.3 patch deadlines | Published June 2024; future-dated requirements mandatory since March 31, 2025 |
| HIPAA Security Rule | US covered entities and business associates handling electronic health data | Risk analysis and safeguards; a proposed rule adds an asset inventory | Proposal from January 2025 not final as of June 2026 |
| SOX Section 404 | US public companies | IT general controls, including change management on financial systems | In force |
| SEC cybersecurity rules | US public companies | Form 8-K Item 1.05 disclosure of material incidents | Since December 18, 2023 |
| CMMC | US defense contractors | Level 2 requires all 110 NIST SP 800-171 requirements | 32 CFR Part 170 since December 2024; contract phase 1 since November 10, 2025 |
| NIS2 | Essential and important entities in the EU | Art. 21 risk measures, supply chain, vulnerability handling; Art. 23 reporting | National transposition varies by member state |
| DORA | EU financial entities | ICT risk management and major incident reporting | Applies since January 17, 2025 |
| EU Cyber Resilience Act | Manufacturers of products with digital elements sold in the EU | Due diligence on components, SBOM, 24-hour reporting | Reporting since September 11, 2026; full application December 11, 2027 |
| FDA §524B | Sponsors of internet-connected medical devices | SBOM, vulnerability monitoring plan, patching | Since March 29, 2023 |
| APRA CPS 234 | Australian banks, insurers, and super funds | Information security capability, including third parties; 72-hour notice | Since July 1, 2019 |
| Privacy laws: GDPR, CCPA, Brazil's LGPD, Canada's PIPEDA, Australia's Notifiable Data Breaches scheme | Organizations processing personal data | Security of processing and breach notification | In force |
Privacy laws matter here for one reason. A dependency that exfiltrates personal data turns a supply chain incident into a notifiable breach, with GDPR's 72-hour clock and fines attached.
The PCI DSS objective quoted at the top of this page is the industry's model in one sentence: keep known vulnerabilities in third-party components from being exploited. It was the right target when the main dependency risk was a stale library. It is the wrong target when attackers publish malware through a maintainer's own release pipeline, with valid provenance, minutes before you install it.
The Cyber Resilience Act points at the fix. It treats any incident that could lead to malicious code in a product as severe, with or without a CVE. Other frameworks should follow it, and ask for evidence that measures outcomes rather than paperwork: what fraction of new dependency versions were analyzed before they ran, and how long a malicious version lived in the environment before anyone knew. OMB's January 2026 memo said M-22-18 imposed "unproven and burdensome software accounting processes that prioritized compliance over genuine security investments." That diagnosis is half right: attestations described process and never checked artifacts. The answer is not less evidence, but evidence about what the code did.
An auditor can verify today that every critical CVE was patched within a month. No major framework asks how many minutes a malicious dependency ran in your CI before someone noticed. That number decides whether you have an incident.
Manifest confusion is a mismatch between an npm package's registry manifest and the package.json in its tarball, two copies that tools trust inconsistently.
A software bill of materials (SBOM) is a machine-readable inventory of the components, versions, and dependency relationships inside a piece of software.
Software composition analysis (SCA) inventories the third-party code in an application and checks each component for vulnerabilities, licenses, and malware.