Annex I of the CRA establishes essential cybersecurity requirements (Part I) and vulnerability handling requirements (Part II) applicable to products with digital elements.
Part I: Cybersecurity requirements
Properties (security by design / default)
- No known exploitable vulnerabilities at time of placing on market.
- Secure default configuration.
- Mechanism to recover or reset in case of incident.
- Confidentiality protection for stored, transmitted, processed data.
- Integrity protection for stored, transmitted, processed data and configurations.
- Authentication and identity management appropriate to risk.
- Authorization controls appropriate to risk.
- Minimize attack surface.
- Reduce impact of incidents including mitigation.
- Provide security-related information to user.
- Log security-relevant events.
- Protect against unauthorized access.
- Provide security updates.
Requirements are risk-based and proportional.
Documentation
Manufacturer maintains technical documentation demonstrating compliance per Annex VII.
Part II: Vulnerability handling requirements
Throughout product support lifetime:
- Identify and document vulnerabilities and components (SBOM).
- Address vulnerabilities without delay; security updates as appropriate.
- Apply effective and regular tests and reviews.
- Disclose information about fixed vulnerabilities once security update available.
- Have coordinated vulnerability disclosure policy.
- Share information about potential vulnerabilities.
- Disseminate security updates effectively.
SBOM emphasis
Vulnerability identification includes identifying components. Software Bill of Materials (SBOM) is the practical mechanism. Standards (SPDX, CycloneDX) widely used.
CVD policy
Coordinated vulnerability disclosure policy required:
- Disclosure channel (typically security@ or vendor-specific).
- Acknowledgement timing.
- Remediation timeline.
- Public disclosure approach.
Support period
Manufacturer commits to support period. Updates provided during support period. Communicated at purchase.
Conformity assessment procedures
Default products
Self-assessment (Module A internal control). Manufacturer's declaration of conformity.
Important products (Annex III)
- Module B (EU-type examination) + Module C (conformity to type) — Annex VI procedures.
- Or full quality assurance Module H — third-party assessment.
Critical products
Mandatory third-party certification through European cybersecurity certification scheme.
CE marking
Affixed after successful conformity assessment.
Reporting obligations (Article 14)
To ENISA:
- Actively exploited vulnerability: notification within 24 hours of becoming aware.
- Severe incident: notification within 24 hours; further updates per ENISA-specified intervals.
To users:
- Notify users about security updates, especially when actively exploited vulnerability addressed.
To national CSIRTs:
- Cooperation expected.
Manufacturer obligations
- Cybersecurity requirements compliance.
- Technical documentation maintenance.
- Conformity assessment.
- Declaration of conformity + CE marking.
- Vulnerability handling.
- Support throughout support period.
- Information to user.
- Cooperation with authorities.
Importer / distributor obligations
- Verify manufacturer compliance.
- Cooperate with market surveillance.
- Cease distribution of non-conforming products.
Open source carve-out
Open source software not commercially supplied is generally outside CRA scope. Commercial supply pulls in scope. Detailed scoping in regulation:
- Open source contributors not in scope when contributing.
- Commercial entities monetizing open source can be in scope.
- "Open source stewards" (organizations supporting open source) have narrow obligations.
Implementation of open source carve-out is complex; guidance maturing.
SRE and AI-agent fit notes
AI-feature software products
AI features integrated into products are part of the product. CRA requirements apply to the integrated whole:
- AI feature vulnerabilities are product vulnerabilities.
- AI feature security updates flow through product support.
- SBOM should identify AI feature components (model versions, libraries).
AI-specific vulnerabilities
CRA-relevant AI vulnerabilities:
- Prompt injection susceptibility.
- Output handling vulnerabilities.
- Model supply chain compromise.
- Inadequate input validation.
Same content as OWASP LLM Top 10 / ATLAS, integrated into product security context.
Stefan-context implementation sketch
- For software products: CRA applies if placed on EU market.
- For SRE / consultancy services: out of CRA scope (services not products).
- For clients producing software products: support CRA compliance work.
See also
- cluster MOC · CRA Controversies
- NIS2 Cluster (operator-side) · SLSA SBOM Cluster (supply chain)