All postsEngineering

Cyber Resilience Act reporting for teams that build software

Since September 2026, CRA manufacturers must report exploited vulnerabilities within 24 hours. Who is in scope, what triggers it, what to have ready.

Johnny Unar8 min read

01

who the reporting duty applies to

Since 11 September 2026, Article 14 of the EU Cyber Resilience Act, Regulation (EU) 2024/2847, has required manufacturers of products with digital elements to report actively exploited vulnerabilities and severe security incidents. The rest of the regulation, including the essential security requirements and conformity assessment, applies from 11 December 2027. Reporting came first, and it's the part that lands on engineering teams, because you can't report what you can't find.

The first question is whether you're in scope at all, and for a lot of web teams the answer is no. The regulation covers "a software or hardware product and its remote data processing solutions". Remote data processing only counts when the manufacturer designed it and the product can't perform one of its functions without it. Recital 11 gives the example of a mobile app that needs an API or database the manufacturer provides; that back end is in scope as part of the app. Recital 12 says the opposite about websites that don't support a product's functionality and about cloud services designed outside a manufacturer's responsibility, and points to the NIS2 Directive as the law that covers SaaS, PaaS and IaaS. A standalone web application used in a browser generally falls outside the CRA, and whether NIS2 applies instead depends on the provider's size and sector. A mobile app, a desktop app, firmware, an installable library sold commercially, or a device and the cloud service it needs to work are CRA products.

The second question is who the manufacturer is. Article 3 defines it as the person who develops a product, or has it developed, and markets it under its own name or trademark. If a studio builds an app that a client publishes under the client's brand, the client is normally the manufacturer and carries the Article 14 duty. The studio's role is whatever the contract makes it: maintaining the code, monitoring dependencies, helping draft the report. Article 69(3) also extends the reporting duty to in-scope products placed on the market before 11 December 2027, so an app shipped in 2022 that's still on the market counts.

This is an engineering reading of the regulation, not legal advice. Whether a particular product is in scope, and who its manufacturer is, depends on facts a lawyer should look at, and the product owner is responsible for that call.

02

what counts as actively exploited

The 24-hour clock doesn't start when a CVE is published, and it doesn't start when a scanner finds a vulnerable package in your image. Article 3 defines an actively exploited vulnerability as one "for which there is reliable evidence that a malicious actor has exploited it in a system without permission of the system owner", and Article 14(1) requires notification of an actively exploited vulnerability contained in your product that you become aware of.

That has practical consequences. A probe in your logs that matches a known exploit pattern isn't automatically reliable evidence of exploitation; a confirmed compromise of a customer's installation through your product is. A listing in CISA's Known Exploited Vulnerabilities catalog means a vulnerability is being exploited somewhere, which is excellent prioritisation data and a strong reason to check your products immediately, but by itself it doesn't establish that your product contains the vulnerability in an exploitable form. The engineering job is to get from "this CVE is exploited in the wild" to "our product version X ships the affected component, and here is what we know about exploitation" quickly enough that, if the answer is yes, you still have most of the 24 hours left.

There's a related duty that's easy to miss. Article 13(6) requires a manufacturer that identifies a vulnerability in an integrated component, including an open-source one, to report it to whoever maintains that component. If your team finds a bug in a dependency while investigating, the maintainers need to hear about it as well as the authorities.

03

the two reporting tracks and their deadlines

Article 14 has two parallel tracks with similar shapes, and all notifications go through ENISA's Single Reporting Platform, which the Commission says has been operational since 11 September 2026. You submit to the CSIRT designated as coordinator in the member state of your main establishment, and the submission is made available to ENISA at the same time.

For an actively exploited vulnerability, Article 14(2) requires an early warning within 24 hours of becoming aware, a vulnerability notification within 72 hours with general information about the product, the nature of the exploit and any mitigations users can apply, and a final report no later than 14 days after a corrective or mitigating measure is available, describing the vulnerability, its severity and impact, any information about the malicious actor and the security update.

For a severe incident having an impact on the security of the product, Article 14(4) sets the same 24-hour early warning, which must say whether the incident is suspected of being caused by unlawful or malicious acts, and the same 72-hour incident notification, followed by a final report within one month of that notification. Article 14(5) treats an incident as severe when it affects the product's ability to protect the availability, authenticity, integrity or confidentiality of sensitive or important data or functions, or when it has led or could lead to malicious code being introduced or executed in the product or in a user's systems. A compromised release pipeline that shipped a poisoned update is the obvious example.

Article 14(8) adds that after becoming aware, the manufacturer has to inform impacted users, and where appropriate all users, about the vulnerability or incident and the mitigations they can apply. Article 64 allows fines of up to 15 million euros or 2.5 percent of worldwide turnover for breaching Articles 13 and 14, although Article 64(10) exempts micro and small enterprises from fines for missing the 24-hour early warning deadline specifically.

04

an inventory you can query in minutes

Every one of those deadlines depends on answering one question fast: which of our shipped product versions contain component X at version Y? If that answer lives in someone's memory, the 24 hours go on archaeology. The fix is to produce a software bill of materials for every release and keep it somewhere queryable.

Syft generates a CycloneDX SBOM from a container image or a directory, and Grype scans an SBOM for known vulnerabilities. Current Grype output includes a knownExploited field populated from CISA KEV, so you can pull out the matches that need attention first:

bash
1syft registry.example.com/field-app-api:2.3.1 -o cyclonedx-json=sbom.cdx.json
2grype sbom:sbom.cdx.json -o json > grype.json
3jq -c '.matches[]
4 | select((.vulnerability.knownExploited // []) | length > 0)
5 | {cve: .vulnerability.id, package: .artifact.name, version: .artifact.version}' grype.json

Run against a stock node:18-alpine image in November 2026, this flagged a KEV-listed CVE in the bundled Node.js 18.20.8 binary. Node 18 has been out of support since April 2025, so there's no upstream fix to ship, and that's the general problem with end-of-life dependencies under a reporting regime: when one of them is exploited, your only options are a migration you should have done earlier or a patch you maintain yourself.

Store the SBOMs per product version, not just per build, so you can answer the question for every version still in customers' hands. A single Postgres table with a GIN index is enough for a small portfolio:

sql
1CREATE TABLE sbom (
2 product text NOT NULL,
3 version text NOT NULL,
4 image_digest text NOT NULL,
5 generated_at timestamptz NOT NULL DEFAULT now(),
6 doc jsonb NOT NULL,
7 PRIMARY KEY (product, version)
8);
9CREATE INDEX sbom_components ON sbom USING gin ((doc -> 'components') jsonb_path_ops);
10
11SELECT s.product, s.version, c ->> 'version' AS component_version, c ->> 'purl' AS purl
12FROM sbom s
13CROSS JOIN LATERAL jsonb_array_elements(s.doc -> 'components') AS c
14WHERE s.doc -> 'components' @> '[{"name": "node"}]'
15 AND c ->> 'name' = 'node';

The containment check (@>) uses the index to find candidate SBOMs, and the lateral join pulls out the exact component version. Your CI job inserts each release's SBOM after the build. Pin the scanners themselves by digest or exact version. By LiteLLM's own account, the Trivy scanner in its CI was the likely entry point for the March 2026 compromise of its PyPI releases, which is exactly the kind of severe incident Article 14(3) is about.

05

what a studio should agree with its clients

If you build apps, firmware or installable software for clients who sell in the EU, the reporting duty sits with them, but most of the capability to meet it sits with whoever runs the build. That's worth settling in writing before an incident rather than during one.

Agree who the manufacturer is for each product and say so in the contract. Agree what the studio does on its side: generating and storing SBOMs per release, watching KEV and advisories for the components those SBOMs list, and telling the client within a set number of hours when a shipped version is affected. Agree who drafts the early warning and the 72-hour notification, who has the Single Reporting Platform login, and how users get notified under Article 14(8), because the person who can push an in-app message or an update is usually on the engineering side. And agree what happens to end-of-life dependencies, since a framework that stopped getting security fixes is a support-period problem the client should decide on with full information.

Then run a tabletop exercise. Pick a real KEV entry from last month, pretend one of your products ships the affected component, and time how long it takes to identify the affected versions and produce a draft early warning with the information Article 14(2)(a) asks for. If that takes more than an hour or two, the gap is almost always the inventory, and the inventory is the part you can fix now.

Thanks for reading.
Johnny Unar

Written by

Johnny Unar

Keep reading

Stop patching.Start building.

Tell us what's slowing you down. We'll bring coffee and opinions.