Last updated:
We recognize Large Language Models (LLMs) as a transformative technology that is fundamentally reshaping how we work, communicate, and exchange ideas. The immense power of AI allows for unprecedented efficiency in conducting research, designing new concepts, and articulating complex thoughts.
However, we acknowledge that LLMs are, by design, statistical engines bound by the inherent limitations of probability. By prioritizing the most statistically probable outcomes, they risk reinforcing "intellectual monocultures" and reflecting the biases of a financialized society. These models often lack common sense, struggle to generate truly innovative knowledge, and can fail when analyzing emerging alternative concepts.
Consequently, the Echoes team utilizes AI according to these strict principles:
Our LLMs are supported by a specialized library consisting of thousands of pages of human-authored material developed over 15 years, alongside a curated knowledge base from trusted institutions in ethical and community banking.
Every word we publish is reviewed and edited by competent team members to ensure it accurately reflects our intended meaning and values.
Much of our AI integration is focused on the graphic design of our documents and digital presence.
At Echoes - and eventually at the European Ethical Bank - all final decisions are made by humans who maintain full responsibility for the outcomes.
During onboarding, applicants upload identity, corporate, and KYC documents. To save time and reduce data-entry errors, we use AI to extract structured fields (company name, registration number, date of incorporation, holder's name, date of birth, document number) from those uploads and pre-fill the relevant forms. This section describes exactly what happens.
Only the documents you upload during onboarding are sent for AI extraction: the business registry excerpt (Step 1), the government-issued ID of the authorised representative (Step 2 - KYC), and any external KYC dossier you choose to upload. We never send your messages, audit logs, financial data, or anything outside the documents you explicitly upload.
Documents are processed by Google's Gemini 2.5 (Pro and Flash) vision models, accessed via the Lovable AI Gateway. The Gateway acts as a routing layer; it does not retain the document contents beyond the duration of the API call. Per the Gateway's terms, the document contents are not used to train Google's or any third party's foundation models.
The AI's only job is to read the document and return structured text fields. The AI does not make any decision: KYC approval, eligibility, credit, risk, or admission decisions are always taken by a human reviewer at Echoes. The extracted fields are a starting point you can correct before submitting any step.
Original files remain in our encrypted Lovable Cloud storage with row-level access control: only you, your assigned regional partner, and Echoes administrators can read them. The structured fields the AI extracted are stored alongside the document for audit. We retain documents for the statutory period required by EU/Italian/Lithuanian AML rules (typically 10 years after the end of the business relationship).
You can ask us to disable AI extraction and review your documents manually. Email info@echoes-holding.eu with the subject "AI opt-out" before submitting Step 1. Your onboarding will take longer but is otherwise unaffected. Consent to AI extraction is recorded in Step 1 with the policy version and timestamp.