Back to glossary

Cybersecurity and Infrastructure Security Agency (CISA)

CISA is the U.S. agency that leads federal civilian cyber defense, issues binding directives to agencies, runs the KEV catalog, and funds the CVE program.

The Cybersecurity and Infrastructure Security Agency (CISA) is the U.S. federal agency that leads civilian cyber defense. CISA issues binding directives to federal civilian agencies, publishes the Known Exploited Vulnerabilities (KEV) catalog, funds the CVE program, and writes voluntary guidance such as the SBOM minimum elements. Outside government, CISA's data shapes how many security teams rank vulnerabilities.

In security, the acronym CISA almost always means the agency. Two other things share it: ISACA's Certified Information Systems Auditor credential, and the Cybersecurity Information Sharing Act of 2015, the law that protects companies that share threat indicators with the government. As of November 2026, Congress has extended that law's expiring provisions only to December 11, 2026, according to the Congressional Research Service.

The most important fact about CISA is how narrow its legal reach is. CISA's directives bind only federal civilian executive branch agencies. Everyone else follows CISA because its data is free, public, and wired into scanners, not because they have to.

Our position: CISA runs the best free public infrastructure for vulnerabilities, and that infrastructure is the wrong instrument for open source malware. Every CISA mechanism that compels action keys on a CVE ID and a remediation deadline. Malicious packages rarely get CVEs. In the five compromises we measure below, CISA's first public action came 4 to 188 days after the first malicious release. Use CISA data to rank CVEs. Do not read CISA's silence about a package as evidence that the package is safe.

What does CISA stand for in cybersecurity?

#

CISA stands for the Cybersecurity and Infrastructure Security Agency, a component of the Department of Homeland Security (DHS). The Cybersecurity and Infrastructure Security Agency Act of 2018 (H.R. 3359) created CISA on November 16, 2018, by renaming and elevating DHS's National Protection and Programs Directorate. CISA's mission covers both cyber and physical risk to critical infrastructure. This entry covers the cyber side, and specifically the parts that touch software and its dependencies.

CISA absorbed the federal government's computer emergency response teams. A CERT, or computer emergency response team, coordinates incident response and vulnerability notices for a defined constituency. US-CERT served federal networks and the public, and ICS-CERT served industrial control systems. CISA retired both names on February 24, 2023. The alerts and advisories that US-CERT used to send now come from CISA directly.

CISA has had no Senate-confirmed director since Jen Easterly left on January 20, 2025. As of November 2026, Nick Andersen has been acting director since February 26, 2026. The nominee, Sean Plankey, asked the White House to withdraw his nomination in April 2026 after 13 months without Senate confirmation. The same Federal News Network report says CISA has lost about a third of its staff since January 2025. The fiscal 2026 budget request had projected the workforce falling from 3,292 employees to 2,324, according to Nextgov/FCW.

What authority does CISA have?

#

CISA's binding authority covers federal civilian agencies and almost nothing else. Under the Federal Information Security Modernization Act (FISMA), CISA issues two kinds of orders to Federal Civilian Executive Branch (FCEB) agencies. A Binding Operational Directive (BOD) sets standing requirements. An Emergency Directive (ED) orders immediate action against a specific threat. Neither binds a private company, a state government, or an open source maintainer.

Three directives define CISA's role in software security:

  • ED 21-01 (December 2020) ordered agencies to disconnect SolarWinds Orion after attackers compromised its build process. It was CISA's first emergency directive aimed at a software supply chain attack. CISA retired it, with nine other emergency directives, in January 2026.
  • BOD 22-01 (November 3, 2021) created the KEV catalog and attached a federal patch deadline to every entry.
  • BOD 26-04 (June 10, 2026) harmonizes BOD 22-01 and BOD 19-02 and replaces fixed deadlines with risk tiers. The tiers depend on internet exposure, KEV status, whether exploitation can be automated, and technical impact (CISA announcement).

The change is visible in the catalog data. All 150 KEV entries added since June 10, 2026 cite BOD 26-04 in their requiredAction field, and none of the 1,617 earlier entries do, as of catalog version 2026.11.03.

CISA reaches past federal networks only indirectly, through contracts and statutes that borrow its rules:

The figure shows three rings. The innermost ring is federal civilian executive branch agencies, which CISA binds directly through BODs, such as the BOD 26-04 deadlines, and EDs, such as ED 21-01's order to disconnect SolarWinds Orion. The middle ring is organizations pulled in by contract or statute. FedRAMP requires cloud providers that sell to agencies to remediate KEV entries on BOD 26-04 timelines, mandatory from December 7, 2026. Federal software suppliers had to sign CISA's Secure Software Development Attestation Form, built on the NIST Secure Software Development Framework (SSDF), until OMB memo M-26-05 rescinded that requirement on January 23, 2026. The form still exists, but each agency now decides whether to ask for it (Mayer Brown). Critical infrastructure operators will owe CISA a report within 72 hours of a covered cyber incident and within 24 hours of a ransom payment under the Cyber Incident Reporting for Critical Infrastructure Act (CIRCIA). As of November 2026 that rule is not in force: CISA sent the final rule to OMB on October 1, 2026, according to Paubox, two years after we covered the proposed CIRCIA reporting rule. The outer ring is everyone else, who uses KEV, Vulnrichment, alerts, the Secure by Design pledge, and the SBOM minimum elements voluntarily.

The rescinded attestation requirement was CISA's strongest lever over commercial software producers. With it gone, CISA's influence on vendors outside FedRAMP runs almost entirely through data and persuasion.

CISA runs the plumbing of vulnerability data

#

CISA's largest effect on software security comes from data it publishes, not orders it issues. Three CISA programs sit underneath many vulnerability scanners.

The CVE program

#

CISA funds the CVE program, which MITRE operates under contract. The program assigns the identifiers that advisory databases, scanners, and SBOM tools use to cross-reference a vulnerability. In April 2025 the MITRE contract came within hours of lapsing before CISA extended it for 11 months, an episode we covered as the CVE contract lapse. In March 2026, CISA and MITRE renegotiated the contract and moved its funding to a protected line in CISA's budget, according to CSO Online. The contract terms remain unpublished. Acting Director Nick Andersen told CSO Online that "the CVE program is fully funded and has continually evolved and modernized." The CVE/NVD pipeline still cannot be the only source of truth, because a single lapsed contract nearly halted the program that coordinates CVE assignment worldwide.

Vulnrichment and the CISA ADP container

#

CISA adds its own analysis to official CVE records through Vulnrichment, launched in May 2024 as NIST's enrichment backlog grew, as we reported when CISA launched Vulnrichment. CISA acts as an Authorized Data Publisher (ADP) in the CVE program. Every CVE it scores gets three Stakeholder-Specific Vulnerability Categorization (SSVC) decision points: Exploitation, Automatable, and Technical Impact. Higher-risk CVEs may also get CWE and CVSS data. The Vulnrichment README is explicit about ownership: "At no point will CISA overwrite the originating CNA's data in the original CNA container in the CVE record."

You can read CISA's container directly from the official CVE list mirror on GitHub. This is the record for the TanStack compromise:

Bash
curl -s https://raw.githubusercontent.com/CVEProject/cvelistV5/main/cves/2026/45xxx/CVE-2026-45321.json \
  | jq '[.containers.adp[] | select(.providerMetadata.shortName=="CISA-ADP") | .metrics[].other]'
JSON
[
  {
    "type": "ssvc",
    "content": {
      "timestamp": "2026-05-27T00:00:00+00:00",
      "options": [
        {
          "Exploitation": "active"
        },
        {
          "Automatable": "no"
        },
        {
          "Technical Impact": "total"
        }
      ],
      "role": "CISA Coordinator",
      "version": "2.0.3",
      "id": "CVE-2026-45321"
    }
  },
  {
    "type": "kev",
    "content": {
      "dateAdded": "2026-05-27",
      "reference": "https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-45321"
    }
  }
]

Two details matter. The CNA that assigned the CVE was GitHub, and CISA's scoring sits beside the CNA's data rather than inside it. The SSVC timestamp and the KEV listing are the same day, May 27, 2026, 16 days after the malicious versions went live. BOD 26-04 uses those same SSVC values to set federal deadlines, which is why their consistency matters later in this entry.

The enrichment effort has also been inefficient. The Commerce Department's Office of Inspector General found at least 21,000 cases between May 2024 and December 2025 in which NIST and CISA duplicated enrichment work, as we reported in our coverage of the Commerce OIG audit of the NVD. CISA stopped producing CPE applicability statements in December 2024, which left NIST as the only federal source of the product mappings scanners match on.

The KEV catalog

#

The CISA KEV catalog lists CVEs with reliable evidence of exploitation in the wild. As of catalog version 2026.11.03, released November 3, 2026, it lists 1,767 CVEs (kev-data). Commercial products dominate: Microsoft accounts for 390 entries, Cisco for 102, and Apple for 95. KEV now steers federal enrichment too. Since April 2026 the National Vulnerability Database (NVD) enriches only CVEs in KEV, software the federal government uses, and software designated critical, as we covered when NIST stopped enriching most CVEs. A CVE outside KEV is now less likely to get CPE and CVSS data from NIST at all.

CISA publishes each of these datasets in a different shape, and each answers a different question:

CISA outputFormat and keyWho must act on itWhat it misses for open source
KEV catalogJSON and CSV, keyed on CVE ID, CC0 licenseFCEB agencies by deadline; FedRAMP cloud providersAnything without a CVE; listings trail exploitation
Vulnrichment ADP containerCVE JSON 5 record, SSVC plus some CWE and CVSSNobody directly; feeds BOD 26-04 tiersNot every CVE is scored; SSVC was designed for flaws, not malware
Alerts and advisoriesWeb pages and email; IT and OT advisories also as CSAF 2.0NobodySupply chain alerts are prose, not version ranges a scanner can match
SBOM minimum elementsGuidance document, format-neutralNobody; voluntary guidanceDescribes what you ship, not whether a component is hostile

Verdict: pull KEV and the CISA ADP container into your scanner as prioritization inputs for reachable CVEs, and subscribe to CISA alerts by email for context. CISA retired the RSS feeds for KEV and alerts on May 12, 2025, as we noted in CISA's RSS retirement, so the KEV JSON on GitHub is the dependable machine feed. None of these outputs detects a malicious package before it installs.

Socket uses the KEV catalog this way. The Socket dashboard shows a "Known Exploits" section with CISA KEV details on affected alerts (changelog), and security policy rules can match on a "CISA KEV" true or false field (rule conditions).

How does CISA respond to open source supply chain attacks?

#

CISA responds to open source supply chain attacks with public alerts and, when a CVE exists, a KEV listing, typically days to months after the malicious code first ran. We compared CISA's first public action against our own first public report for five compromises of packages and GitHub Actions in 2025 and 2026. Day 0 is the first malicious release, taken from npm registry publish timestamps where they exist.

The same data in text:

CompromiseFirst malicious releaseSocket's first public reportCISA alertKEV listing
tj-actions/changed-files (CVE-2025-30066)March 14, 2025March 17, 2025 (day 3)March 18, 2025 (day 4)March 18, 2025 (day 4)
eslint-config-prettier (CVE-2025-54313)July 18, 2025, 15:51 UTCJuly 19, 2025, 00:02 UTC (8 hours)None foundJanuary 22, 2026 (day 188)
Shai-Hulud worm (no CVE)September 14, 2025, 17:58 UTCSeptember 16, 2025 (day 1.7)September 23, 2025 (day 9)Not listed
axios (no CVE)March 31, 2026, 00:21 UTCMarch 31, 2026, 02:52 UTC (2.5 hours)April 20, 2026 (day 20)Not listed
TanStack (CVE-2026-45321)May 11, 2026, 19:20 UTCMay 11, 2026, 20:23 UTC (1 hour)None foundMay 27, 2026 (day 16)

Socket's report times are when our first blog post went live, which trails detection. For axios, our scanner flagged the malicious dependency plain-crypto-js at 00:05 UTC on March 31, 2026, 16 minutes before the first malicious axios version was published, as described in our coverage of the axios compromise. In the TanStack compromise, part of the Mini Shai-Hulud campaign, every one of the 84 malicious package versions was flagged within six minutes of publication. The other three events are the tj-actions/changed-files compromise, the eslint-config-prettier compromise, and Shai-Hulud. The registry still records when the malicious axios versions went out, even though npm removed them:

Bash
curl -s https://registry.npmjs.org/axios \
  | jq '{t1: .time["1.14.1"], t0: .time["0.30.4"], still_listed: (.versions | has("1.14.1"))}'
JSON
{
  "t1": "2026-03-31T00:21:58.168Z",
  "t0": "2026-03-31T01:00:57.285Z",
  "still_listed": false
}

Four patterns stand out.

CISA alerts are guidance, not detection. CISA's September 23, 2025 alert on the self-propagating npm worm reported "over 500 packages" compromised and cited private research, including our analysis of the CrowdStrike packages. One instruction to developers was blunt: "Immediately rotate all developer credentials." That is the right advice, and it arrived nine days after the first malicious publish.

KEV requires a CVE, and malware rarely gets one. Shai-Hulud and axios have no CVEs, so neither can appear in KEV. The xz backdoor did get a CVE, CVE-2024-3094, and CISA issued an alert on March 29, 2024, the day it was disclosed. As of catalog version 2026.11.03, CVE-2024-3094 is still not in KEV. CISA's own ADP container rates it Exploitation=none, consistent with KEV's requirement for evidence of exploitation.

SSVC gives inconsistent answers for malware. Three malicious-code CVEs in KEV received three different SSVC ratings from CISA:

Bash
for id in 2025/30xxx/CVE-2025-30066 2025/54xxx/CVE-2025-54313 2026/45xxx/CVE-2026-45321; do
  curl -s https://raw.githubusercontent.com/CVEProject/cvelistV5/main/cves/$id.json \
  | jq -c '[.cveMetadata.cveId] + [.containers.adp[] | select(.providerMetadata.shortName=="CISA-ADP")
      | .metrics[].other | select(.type=="ssvc") | .content.options[] | to_entries[0] | "\(.key)=\(.value)"]'
done
Plain text
["CVE-2025-30066","Exploitation=active","Automatable=yes","Technical Impact=partial"]
["CVE-2025-54313","Exploitation=active","Automatable=no","Technical Impact=partial"]
["CVE-2026-45321","Exploitation=active","Automatable=no","Technical Impact=total"]

The tj-actions/changed-files payload and the TanStack payload both exposed CI secrets from every pipeline that ran them, with no attacker interaction per victim. CISA rated one Automatable and the other not. SSVC's Automatable decision point asks whether an attacker could exploit a flaw "at will," without reconnaissance or delivery, per the Vulnrichment README. A malicious package inverts that model: the victim performs delivery by running npm install. Under BOD 26-04, Automatable and Technical Impact help set federal deadlines, so a vocabulary built for flaws now decides how fast agencies must respond to malware.

CISA's lockfile advice is incomplete. The Shai-Hulud alert told teams to review package-lock.json and yarn.lock files for affected versions. We argued during the axios incident that a lockfile only covers installs that respect it. Code run through npx, global installs, and fresh CI installs resolve against the live registry and never appear in the project lockfile, as we showed in the hidden blast radius of the axios compromise. A clean lockfile checked after an alert does not prove a team was never exposed.

What are CISA cybersecurity best practices?

#

CISA cybersecurity best practices for software producers come mainly from four places: the Secure by Design principles and pledge, the SBOM minimum elements, the Principles for Package Repository Security, and the remediation steps in incident alerts. Each one helps, and each one has a specific blind spot.

Secure by Design

#

Security by design is CISA's main ask of vendors. CISA's Secure by Design page defines it plainly: "Secure by Design products are those where the security of the customers is a core business requirement." CISA launched the Secure by Design pledge at the RSA Conference in May 2024. The pledge has seven goals: more multi-factor authentication, fewer default passwords, fewer entire classes of vulnerability, more security patch installation, a published vulnerability disclosure policy, transparency in CVE reporting, and better access to security logs. Socket signed the Secure by Design pledge on May 8, 2024.

What the pledge does not cover: none of the seven goals addresses what a vendor's own build installs from a public registry. A vendor can meet every goal and still ship a compromised dependency.

SBOM minimum elements

#

A Software Bill of Materials (SBOM) is a machine-readable inventory of the components in a piece of software. CISA and NSA published the 2026 Minimum Elements for a Software Bill of Materials on July 29, 2026, replacing NTIA's 2021 baseline. The update adds SBOM Author Signature, SBOM Version, and Component Hash Value as new elements, and applies the minimum elements to open source, AI software, and SaaS. It followed the August 2025 draft we covered in CISA's 2025 SBOM guidance update.

What SBOMs do not cover: an SBOM tells you what you shipped, not whether any component is hostile. A component hash proves a file is the file the SBOM names. A hash of a malicious axios tarball verifies just as cleanly as a hash of a clean one. Tools run through npx during a build usually do not appear in the shipped artifact's SBOM at all.

Principles for Package Repository Security

#

CISA co-developed the Principles for Package Repository Security with the OpenSSF and convened registries at its March 2024 open source security summit. Five registries agreed to work toward the principles: the Rust Foundation, the Python Software Foundation, Packagist and Composer, npm, and Maven Central, as we covered in CISA's package registry initiative. The principles define maturity levels, from supporting multi-factor authentication at Level 1 to requiring it for all maintainers at Level 3, along with a vulnerability disclosure policy.

What they do not cover: the principles are voluntary, and the controls they emphasize do not stop a stolen credential. Shai-Hulud spread by stealing npm tokens from developers who had already authenticated. The TanStack versions were published through TanStack's own OIDC trusted publishing binding, so they carried valid provenance. Provenance and signatures are necessary, but they cannot tell you whether the identity that published a release was in control of its own pipeline.

Credential hygiene from CISA alerts

#

CISA's supply chain alerts converge on the same steps: rotate developer credentials, require phishing-resistant MFA on GitHub and npm accounts, audit GitHub Apps and OAuth grants, and block outbound connections to known exfiltration endpoints. These steps limit damage. They all start after the malicious version has been installed.

The gap CISA's model leaves open

#

CISA's machinery assumes a vendor, a flaw, a patch, and a deadline. An open source compromise has an attacker, a version, and minutes. In the cases above, the fastest CISA action came on day 4, and two of the five compromises cannot enter KEV at all, because neither has a CVE.

We do not think CISA should try to become a malware scanner. Detection belongs at install time, at the registry and in the tools that sit between developers and the registry. CISA's highest-leverage moves are the ones it made in 2024: convening registries and pushing them toward safer defaults, because registries can remove a malicious version in minutes and CISA cannot. The move CISA has not made is to publish malicious-package indicators keyed on package names and versions, in a format such as OSV that lockfiles and SBOM tools already understand, instead of forcing malware into a CVE-and-deadline pipeline built for patching appliances. Until that happens, CISA's catalog of exploited vulnerabilities will keep recording package malware weeks after it has done its damage.