Vendors that publish CSAF 2.0 advisories about their own products identify 43.9% of products by model number, 8.7% by Package URL (purl) and 6.0% by CPE. 43.0% of products have a name only. 7.4% of vendor documents state that a product is not affected. The figures come from 907 documents by 36 vendors, changed between 22 November 2024 and 22 November 2026.
Key takeaways
- A tool that matches vendor CSAF advisories by CPE finds 6.0% of the products. Model numbers cover 43.9% and purls cover 8.7% (CSAF Publisher Practice 2026).
- 43.0% of the products in vendor CSAF advisories have a name and no machine-readable identifier. A tool can match those products only by text.
- 7.4% of vendor CSAF documents (67 of 907) state that a product is known not affected. 84.5% name a vendor fix.
- The BSI Warn- und Informationsdienst publishes 77% of the 12,813 documents in the study. Its documents describe other parties' products, so a total across all documents hides what vendors publish.
- Record the supplier model number of each hardware component. In your own advisories, give each product an identifier and state "known not affected" where it applies.
Why does product identification in CSAF matter under the CRA?
A CSAF advisory saves work only when a tool can match its products to your components. Regulation (EU) 2024/2847, the Cyber Resilience Act (CRA), Annex I Part II point (1) tells the manufacturer to identify and document vulnerabilities and components, for example in an SBOM. Article 13(5) tells the manufacturer to use due diligence when it integrates third-party components.
Many manufacturers build machines and devices from supplier components, for example PLCs, gateways, drives and sensors. Their security teams have two questions.
- Can a tool match supplier CSAF advisories to these components automatically?
- What must their own advisories contain, so that their customers' tools can match them?
The OASIS CSAF 2.0 standard, section 3.1.3.3, defines the product_identification_helper for this purpose. The eight fields are cpe, purl, model_numbers, skus, serial_numbers, hashes, sbom_urls and x_generic_uris. The vulnerabilities section defines the product status known_not_affected. A vendor uses this status to state that a vulnerability does not affect a product. This is a VEX (Vulnerability Exploitability eXchange) statement.
What do published CSAF advisories contain today?
The CSAF Publisher Practice 2026 study counted 12,813 CSAF 2.0 documents from 38 providers on the BSI CSAF lister and the CERT@VDE CSAF aggregator. The study keeps vendors that describe their own products apart from organisations that describe other parties' products. The dataset is at cvdportal.com/research/data/csaf-publisher-practice-2026.
| Measure | Vendor PSIRTs | BSI WID | CISA | All |
|---|---|---|---|---|
| Publishers | 36 | 1 | 1 | 38 |
| Documents | 907 | 9,885 | 2,021 | 12,813 |
Documents that state known_not_affected | 7.4% | 0.0% | 1.4% | 0.8% |
| Documents with a fixed status | 58.9% | 0.0% | 10.2% | 5.8% |
| Documents that name a vendor fix | 84.5% | 0.0% | 50.1% | 13.9% |
| Products | 14,956 | 92,840 | 18,708 | 126,504 |
| Products with any identifier | 57.0% | 69.6% | 35.8% | 63.1% |
| Products with a name only | 43.0% | 30.4% | 64.2% | 36.9% |
| Products with a model number | 43.9% | 0.0% | 35.5% | 10.4% |
| Products with a purl | 8.7% | 0.0% | 0.0% | 1.0% |
| Products with a CPE | 6.0% | 69.5% | 0.1% | 51.8% |
The source of the table is the CSAF Publisher Practice 2026 study. A product can have more than one identifier, so the identifier rows do not add up to 100%.
Why does a total across all documents give the wrong picture?
Across all 12,813 documents, 51.8% of products have a CPE and 0.8% of documents state known_not_affected. The BSI Warn- und Informationsdienst (BSI WID) causes most of these totals.
BSI WID published 9,885 documents, which is 77% of the study. 9,883 of them use the csaf_base profile and describe other parties' products. BSI WID gives a CPE to 69.5% of its products and a model number to none. 2 of its 9,885 documents state known_not_affected.
CISA published 2,021 documents about other parties' industrial control systems. CISA gives a model number to 35.5% of its products and a CPE to 0.1%. 64.2% of its products have a name only. For industrial components, look at the vendor and CISA figures, not at the total.
Which identifiers do vendors use?
Vendors use model_numbers much more often than cpe, 43.9% against 6.0%. 43.0% of vendor products have no product_identification_helper field. Purls cover 8.7%. Each of the other five fields appears on 1.9% of vendor products or fewer.
| CSAF 2.0 field | Vendor PSIRTs | CISA | BSI WID | Typical use |
|---|---|---|---|---|
model_numbers | 43.9% | 35.5% | 0.0% | Hardware devices, modules and part numbers |
purl | 8.7% | 0.0% | 0.0% | Software packages, libraries and firmware components |
cpe | 6.0% | 0.1% | 69.5% | Products that have a CPE name |
x_generic_uris | 1.9% | 0.5% | 0.0% | URIs for a vendor's own naming scheme |
skus | 0.9% | 0.1% | 0.0% | Stock-keeping units of product variants |
serial_numbers | 0.2% | 0.0% | 0.0% | Batches or single devices |
hashes | 0.1% | 0.0% | 0.0% | Exact firmware images or files |
sbom_urls | 0.0% | 0.0% | 0.0% | Links to an SBOM |
The source of the percentages is the CSAF Publisher Practice 2026 study. The field definitions come from CSAF 2.0 section 3.1.3.3.
A model number is printed on the device nameplate, the datasheet and the invoice. So a manufacturer can usually find it in its own purchase records.
How often do vendors state "not affected" and a fixed version?
7.4% of vendor documents (67 of 907) state known_not_affected. 84.5% (766 of 907) name a vendor_fix remediation. 58.9% (534 of 907) give a fixed or first_fixed status. 835 vendor documents use the csaf_security_advisory profile and 72 use the csaf_vex profile.
When a supplier gives no known_not_affected status, each customer whose SBOM tool flags the CVE must check the component itself or ask the supplier. The difference between 84.5% and 58.9% has a second effect. Many advisories name the fix in a remediation but give no machine-readable fixed version in product_status.
What should you do with supplier CSAF?
Do not rely on CPE to match industrial suppliers. Record the supplier model number of each hardware component. In vendor advisories, CPE covers 6.0% of products and model numbers cover 43.9%. In CISA advisories, CPE covers 0.1% and model numbers cover 35.5%.
- Record the supplier model number or part number of each PLC, gateway, drive, sensor and embedded module. Keep the supplier's number next to your own part number, so that a
model_numbersfield matches with no manual mapping. - Keep the purl of each software package and library. Keep the CPE where one exists. 69.5% of BSI WID products have a CPE.
- Keep a text match and a manual review queue for products that have a name only. They are 43.0% of vendor products and 64.2% of CISA products.
- Read both
product_statusandremediations. 84.5% of vendor documents name a vendor fix, but 58.9% give a fixed status. - For an important CVE with no VEX status, ask the supplier directly. 7.4% of vendor documents state
known_not_affected, so the other documents do not tell you that a component is unaffected.
What should your own CSAF advisories contain?
Give each product in your product tree an identifier in product_identification_helper. State known_not_affected when a vulnerability does not affect your product. Your customers' tools can then match your advisories with no manual work.
- Give each hardware product, variant and part number a
model_numbersentry. Use the exact designation from your datasheet and order form, so that customers can match it against their purchase records. - Add a
cpeor apurlwhere one exists for your product or its software. CSAF 2.0 section 3.1.3.3 allows more than one field per product, so hardware records and SBOM tools can both match. - Do not publish a product with a name only. 43.0% of vendor products in the study have no identifier, so customer tools must match them by text.
- When you ship a fix, give a
fixedorfirst_fixedstatus together with thevendor_fixremediation. A customer's scanner can then close the finding automatically. - When a public CVE is in a library or chip that your product uses in a configuration that is not affected, publish a
known_not_affectedstatement. Your customers can then record the result for the component in their own vulnerability handling.
How the study was done, and its limits
The study counted 12,813 CSAF 2.0 documents from 38 of the 40 providers on the two lists, read on 22 November 2026. The BSI CSAF lister has 14 providers and the CERT@VDE CSAF aggregator has 26. The study gives results for each source type, never for one vendor.
Two providers are not in the study. Red Hat Product Security publishes one VEX document for each CVE in operating system packages, tens of thousands of files, which would outweigh all other providers. Huawei PSIRT was left out together with Red Hat to limit the size of the download.
The tool csaf_downloader from the gocsaf project downloaded the documents with the option -t 2y. It checks each document against the CSAF JSON schema and its hash. The window is 22 November 2024 to 22 November 2026, by the date in each provider's changes.csv or ROLIE feed. Porta Regulus B.V. ran the study for the CRA Data Institute. The analysis script and the dataset are public under CC BY 4.0, so anyone can run the study again.
The dataset has six limits.
- It contains two years of documents. Older advisories are not counted.
- Each document counts once, at its current version.
- A product that is in more than one document counts once for each document.
- The script counts what the documents state. It does not check if a statement is correct.
- BSI WID and CISA describe other parties' products, so their figures measure the intermediary and not the vendor.
- Products that a CSAF relationship defines, for example firmware installed on a device, are counted apart, because each one combines two products that the main count already holds. Vendor documents define 8,131 such products, and 27.5% of them have an identifier of their own.
How does CVD Portal publish CSAF and VEX?
On the Reporting plan and higher, a public CVD Portal serves a CSAF 2.0 feed at /.well-known/csaf/provider-metadata.json. The feed has directory listings, SHA-256 and SHA-512 hashes, and OpenPGP signatures. Each advisory in the feed gives the product's part numbers, CPE and purl. An admin can also publish the VEX document of each product in the feed.
On 22 November 2026, the feed of the CVD Portal demo portal passed the BSI csaf_checker for the csaf_trusted_provider role. The CSAF export guide gives the setup steps and the feed format.
