Privacy Policy

Data Minimization

Data minimization means that only the strictly necessary information is transmitted to the language model to respond to a query. This means that instead of exposing all customer information (which could include personal data, complete interaction history, or account details), the system only shares a limited and anonymized fragment containing the key elements to generate a useful response.

In this workflow, sensitive data is kept in the RAG retrieval layer, where it is properly protected and managed according to security protocols.

Information Anonymization

Before any data reaches the language model, the system transforms the selected fragment in such a way that any identifying information, such as names, addresses, account numbers, or any other personal data, is removed.

By removing these unique identifiers, it is ensured that the fragment sent to the language model is anonymous, making it impossible to link the information in the generated response to a specific customer. This approach is fundamental to comply with GDPR, as it minimizes the risk of the AI processing personal data.

Access Control and Information Isolation

The database in the RAG is designed so that only authorized personnel or systems can access the complete information. Additionally, the language model does not have access to the original database or the personal data stored in the RAG.

By isolating the Gen-AI Microsystems system from direct access to the complete information, any risk of confidential data exposure is avoided, as the AI generates responses only from anonymous extracts.

GDPR Compliance and Data Protection Policies

This system complies with the privacy by design and by default principles established in the GDPR and other similar regulations. Privacy principles are applied at every stage of processing, ensuring that customer data:

• Is never transmitted to a language model in an identifiable manner.
• Is only used in a strictly necessary way and under anonymization measures.

Additionally, the implementation of this workflow complies with purpose limitation policies, meaning that data is only used for the specific purpose of responding to queries and is not stored or reused for any other purpose.

Protection Against Data Leaks and Breaches

By operating in this manner, the system protects against the risk of data leaks. Since the language model does not receive complete personal information, there is no possibility that this sensitive information will be reflected in the responses or accidentally accessible to others.

Even in the unlikely event that a model response is exposed, it would be impossible to associate it with a specific identity, as the data was previously anonymized.

Transparency and Trust

This anonymization system not only protects customer data but also allows companies to ensure that information handling meets their users' privacy and security expectations.

By informing users about the use of these data anonymization and minimization mechanisms, customer trust can be strengthened, showing a clear commitment to protecting their personal information.

Application in All Our Generative AI Services

All our Generative Artificial Intelligence (Generative AI) services operate under this same security and privacy approach, ensuring the total protection of our customers' sensitive data. Each interaction in our services is designed and executed by applying data minimization and anonymization practices, thus complying with the strictest data privacy requirements, such as GDPR and other data protection regulations worldwide.

We are one of the few companies that implement such an advanced and robust data protection system in their Generative AI services, ensuring that customer data remains secure at every step. This commitment to privacy and information security reflects our industry leadership and dedication to meeting the highest data protection standards.

Our systems operate as follows in each query:

• Extraction and Filtering of Sensitive Information: Each query first passes through a retrieval and information filtering layer that extracts only the relevant data and transforms it into an extract without identifiers or sensitive context. This reduced fragment is the only data sent to the Generative AI model, ensuring that all customer personal information remains confidential and is not accessible or processed beyond the necessary limits.
• Effective Anonymization in All Services: In all our Generative AI services, the system consistently anonymizes the information before sending it to the language model. This means that any identifiable data (names, addresses, account numbers, etc.) is removed so that the model only works with fragments devoid of personal context.
• Strict Isolation and Access Control: The architecture of all our services ensures that AI models do not have direct access to the customer's complete database but operate only with the necessary and already anonymized fragments. This eliminates any possibility of unauthorized access to confidential information or data leaks in the generated responses.
• Comprehensive Compliance with Global Regulations: All our services are designed to comply with the privacy by design and by default principles established in GDPR and global data privacy regulations. Our data processing workflows are built to ensure that only the minimum necessary data is transmitted and processed for each case, avoiding any unauthorized or improper use of sensitive information.
• Protection Against Data Breaches and Response Security: The data minimization and anonymization we implement in all our Generative AI services ensure that responses never contain sensitive data or specific customer identifiers. This protects against possible data leaks or breaches, offering a secure and reliable service.