1. What AI System We Use
CEFRHub uses a large language model (LLM) from Google, through Google Cloud Vertex AI (Google LLC), to analyze linguistic content and generate CEFR level assessments.
- System: Google Cloud Vertex AI (Google LLC)
- Provider: Google LLC
- Type: Large Language Model (LLM)
- Our role: Deployer — we use the model, we did not build it
CEFRHub is a deployer of a third-party AI model, not a developer. Google LLC is responsible for the model itself. CEFRHub is responsible for how it is used.
2. How We Use It
The AI performs a single, clearly defined task:
- Analyzing written text or spoken language submitted by the user
- Specifying a CEFR level (A1 → C2) based on linguistic criteria from the Common European Framework of Reference
- Generating a structured linguistic analysis (grammar, vocabulary, coherence, fluency)
- Providing personalized improvement recommendations
- Profiling users or building behavioral models
- Advertising or marketing targeting
- Biometric identification or emotion recognition
- Any purpose beyond linguistic CEFR assessment
3. Data Sent to the AI
Data transmitted:
- Your submitted text (written evaluation)
- Spoken language (oral evaluation)
- The target evaluation language (French or English)
- The evaluation mode (written / oral)
- Your name, email address, or any account identifier
- Your payment information
- Your evaluation history or previous results
- Any data that could directly identify you
Your results and the text excerpts we keep are stored in Canada (Montréal); audio recordings are never stored. Linguistic analysis is performed by Google Cloud Vertex AI in the Montréal region (northamerica-northeast1). Your content is not used to train Google's models (Google Cloud terms applicable to Vertex AI).
4. Limitations & Accuracy
AI-based CEFR assessment has inherent limitations you should be aware of:
- Indicative results only: CEFRHub results are not equivalent to official certified examinations (DELF, Cambridge, IELTS, TOEFL, etc.)
- Sample length matters: Texts under 150 words or audio under 60 seconds produce less reliable results
- Borderline levels: The AI may under- or over-estimate levels at boundaries (e.g., B1/B2)
- Language variety: Accuracy may vary by accent (oral) or dialect
CEFRHub does not issue official language certifications. Results are an educational and self-assessment tool.
5. No Automated Decision-Making
No decision with legal effect or significant impact on you is made solely by AI output (see GDPR Article 22).
Your CEFR result is an informational and educational tool. Any consequential decision — hiring or admission — is made by a human (employer, institution, authority), not by CEFRHub's system.
Results are generated automatically by AI. They are indicative and intended to support, not replace, the judgment of a teacher or assessor.
6. Human Oversight
We maintain the following mechanisms for human control and correction:
- "Report an error" button on every evaluation result page — reviewed by our team
- Manual review available upon request at support@cefrhub.ai
- Full deletion of your evaluation history available at any time from your profile settings
- Regular prompt and model configuration review to correct systematic biases
7. Our Role & the EU AI Act
This system is an assessment support tool. Its classification under the EU AI Act is under legal review.
What we do as a deployer (EU AI Act):
- Users are informed that AI is used in the assessment process
- Human oversight mechanism is in place
- No prohibited AI practices are used (no manipulation, no subliminal techniques)
- No biometric categorization or emotion recognition
- No deceptive AI interactions (the system never claims to be human)
Canadian context
CEFRHub Technologies Inc. is incorporated federally in Canada. Bill C-27, which contained the Artificial Intelligence and Data Act (AIDA), died on the Order Paper when Parliament was prorogued on January 6, 2025: there is currently no federal AI-specific law in Canada. We created this transparency page on our own initiative.
CEFRHub has completed a Privacy Impact Assessment (PIA / EFVP) under Quebec's Law 25 prior to deploying its AI analysis system. This document is available upon request for institutional clients.
8. Questions & Contact
For any question about our use of AI, to request a manual review, or to exercise your rights:
Privacy Officer: dpo@cefrhub.ai
Technical support & manual review requests: support@cefrhub.ai