AI Transparency
(Changes of 16 November 2026: clauses added or updated: "Copyright policy" and "Machine-readable marking and small models". Type: policy and regulatory alignment, AI Act Art. 50 and Art. 53.)
Art. 1 · Executive summary
YecoAI designs and operates AI systems with human oversight, transparency, and fundamental-rights protection in mind, toward the applicable obligations of Regulation (EU) 2024/1689 (EU AI Act) as they become enforceable for our role (provider and/or deployer). This page covers classification, transparency, oversight, individual rights, risk management, documentation, and governance. Key timelines: prohibited practices from Feb 2025; GPAI obligations from Aug 2025; high-risk obligations from Aug 2026–Aug 2027 as applicable.
Art. 2 · Classification
DeskNexo — AI Support Agent. Risk class: Limited Risk (Art. 50 transparency). Automates WHMCS support tickets via AI analysis and response generation. Interacts with users and generates content. Transparency obligations applied; human oversight implemented.
Other AI projects: research and development (including Anti-Loop Layer and related tooling). Upon deployment, each undergoes risk assessment and classification. The Yeco LLM Series models (Yeco-PII, Ender-1, Yeco Flash Mini-IT) are models made available to our own products and, where openly published, subject to applicable GPAI or open-source rules. Deployment-specific classifications are published as systems reach production.
Emotion recognition & biometrics: YecoAI does not currently deploy emotion recognition or biometric categorization systems. If that changes, users will be informed at interaction per Art. 50.
Art. 3 · Transparency obligations (Article 50)
Users are explicitly informed when interacting with an AI system. DeskNexo notice shown with AI-generated responses: "This response was generated by an AI assistant. Reviewed by human staff."
- Clear "AI" label or equivalent notice
- Notice before/during the interaction
- Clear statement you are speaking with AI, not a human
- Option to request human intervention
AI-generated content marking: text carries visible disclaimers and metadata markers where feasible; email signatures indicate AI assistance when applicable. Synthetic media include disclosure and watermarks where feasible.
Machine-readable marking and small models
Article 50(2) of the AI Act requires synthetic content to be marked in a machine-readable format, taking into account the specificities and limitations of different content types, the costs of implementation, and the generally acknowledged state of the art.
For our small models we do not apply a statistical watermark in the token stream: altering the token probability distribution to embed one significantly degrades the quality and coherence of the text, and on short outputs such as chat replies it cannot be detected reliably. We therefore state this technical limitation openly.
To compensate at system level: responses in our interfaces show users a visible notice that the content is AI-generated (Art. 50(1)); API responses include metadata indicating AI generation, for example an "ai_generated": true field and the model identifier. For exported files we use C2PA metadata where the tooling supports it.
What is NOT marked as AI: human-authored company content, manually written documentation, and non-AI product UI do not require Art. 50 AI labels.
Art. 4 · Human oversight
- Human operators can override AI decisions via control interfaces
- Users may request human review ("Request Human Agent" where available, or by contacting support)
- Automatic escalation when confidence is low, policy flags trigger, or the user requests a human
Art. 5 · Rights of individuals
Right to know you are interacting with AI; right to human intervention/review; right to explanation of significant AI-assisted decisions where applicable; GDPR rights for personal data processed by AI systems. Exercise: [email protected] — Subject: "AI Transparency - Rights Request" (response within 30 days).
Art. 6 · Training data and product use
This section explains which data may be used to train YecoAI's AI models, where it comes from, how it is prepared, and which rights individuals have.
Which data may be used for training
We use for training only data described in the privacy notice of the product it comes from, in compliance with applicable law. In particular:
- we use product data collected under the terms of the product itself, or data for which explicit consent (opt-in) was given;
- we do not use for training data from private contacts, data sent through the contact form, pitch deck requests, or analytics.
How data is prepared before training
Before any data enters a dataset, it goes through two steps.
- Product first-level filter. The product applies a filter that, by default, masks basic personal data. The filter is designed to reduce the presence of personal data in the content that continues through the process.
- Yeco-PII. The content then passes through our personal data anonymization model, Yeco-PII, designed to remove personal data and the context that can identify a person. The process is built so that no identifying data is intended for training datasets after this step.
On Yeco-PII performance we report only published, measured results: average F1 of 0.966 across six European languages for Yeco-PII v1.0, against 0.776 for OpenAI Privacy Filter (1.5B), under the historical v1.0 protocol (17-tag gold).
No automated filter is perfect. To reduce residual risk, where applicable we carry out human review of data before it is used in datasets.
Legal basis and retention
Processing of data for training is based on Article 6 GDPR. Where the law requires it, the legal basis is the consent of the data subject (Article 6(1)(a)).
Training datasets are kept for no longer than strictly necessary for training purposes, and are deleted or anonymized at the end of that period.
Data already used in a trained model
If a piece of data has already been used to train a model, that individual data point may not be extractable from the model itself. In that case we handle the request by excluding the data from future training datasets and applying the retention limits that apply to the data we hold.
Request to delete your data
Anyone whose data is processed by YecoAI can request deletion of the data we hold about them, as provided by the GDPR. Recommended channel: write to [email protected] with the subject "AI Training - Data Removal Request". The subject is not mandatory; it only speeds up handling. Any written request sent to [email protected] or to the PEC [email protected] is also valid.
We reply within one month of receiving the request, as provided by Article 12(3) GDPR. The rights provided by the GDPR are described in the privacy notice.
Data controller
The data controller is YecoAI di Marco Nasi, VAT 04219920040. For any question about the processing of data for training: [email protected].
Art. 7 · Copyright policy
This section describes how YecoAI respects copyright when collecting, selecting and using data to train its models, in line with Article 53(1)(c) of the AI Act and Article 4 of Directive (EU) 2019/790 on copyright in the Digital Single Market.
Respecting rights reservations (opt-out)
When collecting and selecting datasets, including news crawls, we respect machine-readable opt-out protocols: robots.txt, the noai and noimageai meta tags, and the tdm-reservation protocol, as provided by Article 4(3) of Directive (EU) 2019/790.
License verification
We use only open-source datasets with licenses compatible with commercial use (for example CC BY 4.0, MIT, Apache 2.0), or proprietary or synthetic data. We discard material subject to Non-Commercial (NC) restrictions.
Measures against reproduction
Our models are trained and configured to avoid memorizing and verbatim (word-for-word) reproduction of protected works.
Reports from rights holders
Publishers and rights holders can write to [email protected] with the subject "Copyright - [Topic]" for reports, verification requests or opt-outs.
Art. 8 · Risk management, documentation, conformity
Risk management covers design, deployment, monitoring, and incident handling. Technical documentation proportionate to system risk is available to competent authorities and, where applicable, to users: [email protected] — Subject: "AI Act - Technical Documentation". Conformity assessments are carried out for any system that may qualify as high-risk before such obligations apply.
Art. 9 · Governance
Accountable entity: YecoAI di Marco Nasi (VAT 04219920040). Continuous regulatory monitoring; staff training on AI Act, ethics, and compliance procedures.
Art. 10 · Contact
[email protected] — Subject: "AI Act - [Your Topic]" or "AI Transparency - [Your Topic]". Legal representative: YecoAI di Marco Nasi · VAT 04219920040. Complaint authority (Italy): Garante — www.garanteprivacy.it