Method
What a fail-closed check sees from pages already published
What a fail-closed check sees from pages a company already published. Named organisations are examples, not customers, and not legal findings.
Named organisations are not customers. The examples below are reconstructed from cited public pages only. This is not an accusation of non-compliance, not a notified-body assessment, and not a claim that they use this product. Annex III high-risk duties apply 2 December 2027. Article 50 has applied since 2 August 2026.
How the check works
Same rules as a customer system. Public pages often yield a model card. They rarely yield an approval timestamp or a dataset license.
- 1
Read what is already public
Press notes, product pages, and AI Act stance pages the organisation published. Every example on this site cites a URL.
- 2
Rebuild only what those pages support
A model card can often be reconstructed. An approval timestamp or dataset license usually cannot — public pages rarely publish them.
- 3
Run the same fail-closed checks
The same promotion and dataset checks used on customer systems. If a required field is missing, the check does not pass. We do not fill gaps.
What we do not infer
- We do not invent an approval timestamp or dataset license.
- We do not treat a published legal stance as a release record.
- We do not call a company non-compliant or illegal.
- Named organisations are examples, not customers.
Four public examples
Quotes and URLs come from pages those firms already published. A BLOCKED example is not a finding that the organisation is illegal.
Heidelberg, Germany
Getsafe GmbH
Press dated 5 June 2025: AI agents already handle claims, advice, and contract close.
Our AI agents already handle a large part of claims settlement, consulting, and contract conclusion.
Public pages do not publish an approval timestamp or dataset license, so the check does not pass.
Source: Getsafe relies on AI instead of licensesDüsseldorf, Germany
auxmoney GmbH
COO writing, December 2025: machine learning in credit-risk, >90% automation.
Das Ergebnis ist ein durchgängig digitaler Prozess mit über 90 Prozent End-to-End-Automatisierungsgrad. Machine Learning ist seit Jahren fester Bestandteil unserer Arbeit.
Public pages do not publish an approval timestamp or dataset license, so the check does not pass.
Source: Kreditwürdigkeit mit KI im Detail verstehenBerlin, Germany
softgarden ltd.
Product pages: AI Matching highlights applicants; a recruiter decides.
On request, the feature compares the requirements of the job posting with the content of incoming applications … AI then highlights promising candidates.
No public approval or deploy timestamp. Ranking applicants is employment-shaped — this is not a legal classification.
Source: AI recruitment solutions — softgardenMunich, Germany
Retorio GmbH
Publishes an AI Act stance and describes HR selection / development.
Retorio is used in HR to support companies in the selection and development of employees.
A published legal stance is not the same as an approval record or eval pack. This example does not adjudicate Article 5.
Source: The AI Act — Retorio
Questions? See the FAQ and Disclaimer.
Ready to see your release gate?
Request a demo for one named AI system. We follow up with a written quote. This is not a legal certificate.