← Back to home

AI transparency

Last updated: August 6, 2026

Integral is an AI system. This page explains, in plain language, what that means when you use it — what it is good at, how it fails, and where a human decision is always required.

We publish this because anyone relying on a system like this should be able to judge how far to trust it, and because the EU AI Act requires providers and deployers of AI systems to be transparent about exactly that.

1. You are working with an AI system

The assistant, the reports, the workflow steps and the drafted e-mails are generated by large language models. Text you read in those places was written by software, not by a person at MeteorIT.

Generated files and reports carry a note recording that they were produced by Integral and when.

2. What it can get wrong

  • It can state something false with complete confidence. Fluency is not accuracy.
  • It can misread a document, a table or a ticket, especially where formatting is unusual.
  • It can miss information that exists in a connected system but was not retrieved for that particular question.
  • It reflects what is in your connected systems — if the source data is wrong or out of date, the answer will be too.
  • Repeating the same question can produce differently worded, occasionally differently reasoned, answers.

3. Where a human stays in the loop

Actions that change something outside the Platform — sending an e-mail, creating or updating a record in a connected system — are prepared as a draft and executed only after a person reviews and approves them. That gate is deliberate, and it is the main control against a wrong answer becoming a wrong action.

Read-only answers, reports and summaries are not gated in the same way, so treat them as material to check rather than conclusions to act on unexamined.

4. What Integral must not be used for

The Platform is a business productivity tool. It is not designed, tested or offered for uses the EU AI Act classifies as high-risk, and the Terms of Service prohibit them. In particular, do not use Integral to:

  • screen, rank, shortlist or evaluate job applicants, or make or materially inform hiring, promotion or termination decisions;
  • monitor or evaluate the performance or behaviour of workers;
  • assess creditworthiness, set a credit score, or decide eligibility for essential private or public services;
  • assess risk in insurance pricing for life or health cover;
  • perform biometric identification, emotion inference or categorisation of people;
  • support law enforcement, migration, asylum or border control decisions;
  • provide medical, legal or financial advice to an individual as if it were a professional opinion.

If you need a system for any of those purposes, it must be built and assessed as a high-risk AI system under the AI Act. Integral has not been, and using it that way would put both of us outside the law.

5. Human oversight is your organisation’s responsibility too

Under the AI Act, an organisation deploying an AI system must ensure the people using it understand it well enough to use it sensibly. Point your team at this page, and make sure whoever approves drafted actions knows they are approving a machine-generated proposal rather than confirming a colleague’s work.

6. Questions

For questions about how AI is used on the Platform, or to report an output that caused a problem, write to ai@meteorit.rs. We want to hear about the failures.