It is simply not enough to plug two open-source scanners into the CI/CD pipeline.
1. Why scanners & SBOM tools alone are not enough
The Cyber Resilience Act (CRA) requires processes, not just static reports. A scanner does find known syntax patterns or known CVEs in open-source libraries, but:
Logic errors & hallucinations: No scanner sees whether the AI has set up a tenant separation (multi-tenancy) incorrectly, so that user A sees the data of user B.
Risk analysis obligation: The CRA requires a documented risk analysis before and during development. A CI/CD runner, however, only documents the result, not the design.
Vulnerability handling: If a vulnerability becomes known tomorrow, the CRA requires a process to report actively exploited vulnerabilities (early warning within 24 hours, notification within 72 hours) and to fix them without delay through security updates. The scanner only says that something is there – you have to solve it yourself.
2. How long does it take to set up such a compliance pipeline?
If a company wants to set this up seriously from scratch with its own DevSecOps engineers:
Initial effort: 2 to 6 months (including configuration of SAST, DAST, software supply chain management, policy-as-code and the set-up of reporting dashboards).
Ongoing effort: At least 0.5 to 1 FTE (full-time equivalent) to maintain the rulesets, evaluate false positives and update the tooling landscape.
You are essentially building your own small security platform within your company.
3. Lock-in effect: Are you tying yourself down?
Here one has to distinguish between tool lock-in and process lock-in:
With your own pipeline with free AI agents:
Code level: No lock-in. The code belongs to you and runs anywhere (e.g. on any Linux server).
Compliance level: Massive tool & knowledge lock-in. If you wire ten different tools together, you are extremely dependent on this specific pipeline. If the DevSecOps team leaves, nobody understands any more why the pipeline turns red.
With platforms like 1toX-APA:
Compliance level: No effort for pipelines, but framework lock-in. You are bound to the architecture and ecosystem of the provider.
4. Do I still remain fully responsible?
In short: Yes, absolutely.
Legally, the principle of liability of whoever places the product on the market applies. Regardless of whether you use free AI agents with your own pipeline or 1toX-APA: in the end, your company's name is on the EU declaration of conformity (the CE mark for the software). You can never transfer the ultimate legal responsibility to a tool, a framework or an AI.
The decisive difference lies in the liability and exoneration risk:
Grossly negligent (free AI agents without guardrails): If a vulnerability is exploited and it turns out that unseen AI code was put live without documentation, the full CRA fines apply (up to €15 million or 2.5% of worldwide annual turnover).
Exoneration through provability (with framework/hardened process): If a vulnerability occurs despite all due care, you can prove to the authorities without gaps that you developed according to the current state of the art (risk analysis, SBOM, automated barriers). This protects against ruin and accusations of gross negligence.