Any content you send us as part of an inquiry is treated as strictly confidential and used exclusively to prepare your report. It is never shared with other clients or competitors. Your data is processed by established AI infrastructure providers, with a Zero Data Retention requirement - see the sections below for exactly what that means.
Last updated: 23 October 2026.
The content of your request is not shared with other clients or competitors. It is sent to AI providers through OpenRouter, with the requirement that only providers with a Zero Data Retention policy be used, with the exceptions described on the Confidentiality page.
We send calls to AI providers with a zero-retention requirement (with documented exceptions).
The data collection opt-out is turned on: we require providers not to use your data to train their models.
Expert Consilium uses the data you submit only to process the requested consultation and prepare the final report, subject to the applicable privacy and retention rules. Analysis is generated with the help of established AI infrastructure providers, the same standard used by thousands of companies worldwide in their day-to-day operations.
Model requests during consultation processing are routed only through endpoints that satisfy the active Expert Consilium confidentiality policy, including mandatory Zero Data Retention (ZDR), with the exceptions listed below. Compliance is verified at the specific provider-endpoint level, not assumed merely from a model name or routing platform.
ZDR concerns provider-side retention of prompts and model responses: your data is not retained long-term after the response is generated and is not used to train models. It does not mean that no operational data exists anywhere - Expert Consilium may process or retain the minimum application records and technical metadata required to operate the service. Exceptions to ZDR are: the inappropriate-content check (OpenAI Moderation), the model that searches the internet, and models that run at Google (including Gemini), where the provider may, under its own rules, retain requests for safety monitoring and abuse prevention. The duration and terms of that retention are set by the provider, and we do not control them.
Before your question goes to the AI models, our system automatically replaces with placeholders the data it can reliably recognize: email addresses, phone numbers, IBANs and bank accounts, payment card numbers and personal ID, social security and passport numbers of many countries (when the format is valid or a label such as passport no or ID number precedes the number), as well as tax and company registration numbers when the number carries a label (for example PIB, MB or VAT). We do not automatically recognize ID or passport numbers of other countries. The models work with the placeholders, and in the report you see your real data. We do not automatically recognize names of people and companies, addresses, amounts or dates, because wrong masking would spoil the analysis, so we do not claim to remove all identifying data. If you do not want a name sent to the models, put it in double square brackets, for example [[Acme Ltd.]], and it will be replaced with a placeholder.
The content of your inquiry and report is processed automatically, in line with the ZDR policy described above. Apart from the AI providers that process the request, no external collaborator has access. Client reports and internal consultation records are not made available to other clients.
The request for a report is sent through OpenRouter (USA) to the AI model makers. The model makers we currently use are Anthropic, Google, OpenAI, xAI, Meta and Perplexity; all are headquartered in the United States, and the actual place where a given model runs may differ and is decided by the routing provider. The list may change, and the list shown here is the one in force. The application server and backups are in Germany (Hetzner). Data is therefore also processed outside Serbia, including in the USA. The transfer is made only to provide the service you ordered, with masking of identifying data and the Zero Data Retention requirement described above. We determine the basis for the transfer per provider (a data processing agreement with standard contractual clauses or another basis the law allows) and we do not claim it is confirmed for every provider until it has been verified; we provide the list with the basis per provider on request.
One of the AI models in the analysis searches public sources on the internet. Requests to that model use the same ZDR routing as the others, and the text of your question, masked in the same way, may be used as a search query. For that model, ZDR applies to request routing; the search itself is carried out by the model provider under its own data retention rules. If you do not want a piece of data to appear in the query, put it in double square brackets.
Before analysis, your question goes through a check for inappropriate content. A local word list runs on our own servers. When a broader check is needed, the masked text is sent to the OpenAI Moderation API, a dedicated service separate from the AI models that write the report. Per OpenAI's published policy, that service does not use data to train models; any retention period is set by OpenAI under its own terms.
We keep records of requests and reports for delivery, complaints and protection against misuse; we delete them at your request.
SR
EN