1What does this mean?

UNIIOS uses sovereign European language models (LLMs) to assist with certain tasks: CV analysis, automatic translation, suggestion of jobs and skills from the reference list, drafting assistance for job offers and Job Dating descriptions.

No automated decision is taken by AI alone. Every produced suggestion is editable, modifiable or rejectable by the user concerned or a human administrator before publication, saving or matching. UNIIOS keeps a human in the loop at every step, in compliance with article 14 of the AI Act (human oversight) and article 22 of the GDPR (automated individual decisions).

2Classification under the AI Act

CV analysis and candidate/job matching are classified as high-risk AI systems under Annex III, point 4 of Regulation (EU) 2024/1689 (employment, workers' management and access to self-employment, including recruitment and screening of applications).

On this basis, UNIIOS, as a deployer within the meaning of article 3 §4 of the AI Act, applies all obligations set out in article 26 of the regulation, applicable from 2 August 2026.

Multilingual translation and drafting assistance features are classified as limited-risk and are subject to the transparency obligations of article 50 of the AI Act.

3How it works in practice

Four use cases, one rule: AI proposes, you decide. You stay in control at every step.

In short: AI accelerates repetitive work, you keep control of every decision concerning you.

4LLM providers, European sovereignty

UNIIOS relies exclusively on European models hosted in France:

Why this choice. Scaleway is a French cloud provider, subject to European law, outside the scope of the US Cloud Act. Mistral AI is a model developed by a French company. Inference is performed in a data center located exclusively in France.

5Data policy, Zero retention

UNIIOS applies a zero retention policy on content processed by AI:

Our inference provider's commitment. Scaleway applies a Zero Data Retention policy on its Generative APIs: models are stateless (prompts and completions are not retained after processing), data is not accessible to other clients, model creators or third parties, and is not used to improve models. All traffic is encrypted in transit (TLS).

What we track. To ensure regulatory traceability (AI Act article 26 §6, see section 6), we keep only anonymized technical metadata: model identifier called, timestamp, HTTP status, latency, input/output token count, internal request identifier. No prompt or response content is included in this metadata.

6Technical log retention

This 6-month period corresponds to the minimum obligation under article 26 §6 of the AI Act for deployers of high-risk systems, not article 19 (which addresses providers of AI systems, not deployers).

7Human oversight and internal controls

In accordance with articles 14 and 26 of the AI Act:

• Systematic human validation: no AI result produces an effect (publication, matching, saving) without explicit human validation.
• Trained personnel: our oversight teams are trained in AI literacy in accordance with article 4 of the AI Act (applicable since 2 February 2025).
• Global kill switch: UNIIOS administrators have a kill switch allowing them to instantly disable all AI calls in case of anomaly, drift or incident.
• Continuous monitoring: monitoring of error rates, latencies and user reports; systematic escalation of serious incidents to the provider (article 26 §5) and, where appropriate, to the competent market surveillance authority (article 73).
• Compliance with provider instructions: use of models in accordance with the user manuals (article 26 §1).

8Your rights

Under the AI Act and GDPR, you may at any time:

• Be informed that you are interacting with an AI or that content has been generated by an AI (article 50 AI Act);
• Request a clear and meaningful explanation about the role of AI in a decision concerning you and the main elements of that decision (article 86 AI Act);
• Access the list of AI processing performed on your data;
• Rectify AI-extracted data that would be inaccurate;
• Object to automated processing;
• Lodge a complaint with the competent market surveillance authority (article 85 AI Act) or with the CNIL.

9Candidate and employer information

In accordance with articles 26 §7 and 50 of the AI Act:

• Candidates are informed, before any CV upload, that their application may be analyzed by an AI system, and that any matching with an offer remains validated by themselves and the recruiter.
• Employer users of the platform commit, as ultimate deployers of the produced results, to inform their candidates of the use of AI in their pre-selection process.

10Regulatory timeline

11Updates to this page

This page is updated at every significant change to our AI architecture, applicable law, or transparency commitments. Version history is kept on request at dpo@uniios.com.