Log inFree trial

The EU AI Act: how Neurally complies

The European regulation on artificial intelligence, as amended by Regulation (EU) 2026/1744. Here are the articles that concern an AI agent like Neurally: what they require, what Neurally does and what is left to your company.

This page explains Neurally’s position. It is not legal advice.
Read the RegulationRequest a demo

Three roles, three different responsibilities.

The AI Act distinguishes between those who develop the models, those who build the system that uses them and those who use it in a company (Art. 3). Each has its own obligations.

  1. Develops general-purpose AI models

    Model provider

    Documents the models, informs those who integrate them and publishes a summary of the training data. These are the obligations of Articles 53 and 55.
  2. AI system provider

    Neurally

    Develops the AI agent and places it on the market under its own name. It integrates third-party models without retraining them: under the Regulation it is a downstream provider, responsible for the system’s obligations.
  3. Deployer

    Your company

    Uses Neurally under its own authority, for professional purposes. It has lighter obligations of its own, listed further down.

The dates that matter, updated for 2026.

The Regulation applies in stages. Regulation (EU) 2026/1744, the so-called Digital Omnibus, in force since 27 July 2026, moved the deadlines for high-risk systems and added new prohibitions.

  1. 1 Aug 2024Entry into force
  2. 2 Feb 2025Definitions, AI literacy, prohibited practices
  3. 2 Aug 2025General-purpose models, governance, penalties
  4. 2 Aug 2026General application, including the transparency rules of Art. 50
  5. 2 Dec 2026New prohibitions and marking for systems already on the market
  6. 2 Dec 2027High-risk systems in Annex III
  7. 2 Aug 2028High-risk systems in Annex I
The Commission’s official timeline

What the Regulation requires, what Neurally does.

For each article: who it applies to, from when, what it requires and how Neurally responds. “Voluntary” marks the high-risk requirements that Neurally applies even though it is not obliged to.

Applies toNeurallyYour companyBothVoluntaryModel providers

  1. Art. 3

    DefinitionsBothfrom 2 Feb 2025Official text

    What it requires

    It distinguishes the provider of an AI system, who develops it and places it on the market under its own name, from the deployer, who uses it under its own authority.

    What Neurally does

    Neurally is the provider of the system and integrates third-party models as a downstream provider. Your company is the deployer.

  2. Art. 4

    AI literacyBothfrom 2 Feb 2025Official text

    What it requires

    Providers and deployers take measures to support the AI literacy of the people who use AI on their behalf, taking the context of use into account.

    What Neurally does

    The Help Centre answers questions about using Neurally from the official documentation, with video tutorials in the same panel. Your company only needs to keep a record of the training it has done: according to the Commission, no certificate is required.

  3. Art. 5

    Prohibited AI practicesBothfrom 2 Feb 2025Official text

    What it requires

    Among other things, it prohibits manipulation, exploitation of vulnerabilities, social scoring, emotion recognition at work and in education, and biometric categorisation.

    What Neurally does

    Neurally searches company documents and answers questions about them: it is not designed for any of the prohibited practices and does not generate images, audio or video.

  4. Art. 6

    High-risk systems and Annex IIIYour companyfrom 2 Dec 2027Official text

    What it requires

    A system is high-risk when it is used in the areas of Annex III: biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice.

    What Neurally does

    Neurally’s functions do not in themselves fall within these areas. If you want to use it to make decisions in one of them, for example to screen CVs, the classification changes: talk to us first.

  5. Art. 10

    Data and data governanceVoluntaryhigh-riskOfficial text

    What it requires

    For high-risk systems: quality training, validation and testing data, with documented governance practices.

    What Neurally does

    Customer documents are used only to answer the people who have access to the knowledge that contains them, and are not used to train the models.

  6. Art. 12–14

    Record-keeping, transparency, human oversightVoluntaryhigh-riskOfficial text

    What it requires

    For high-risk systems: automatic logging of events, clear instructions for deployers, and people able to oversee the system and stop it.

    What Neurally does

    Neurally applies them even though it is not high-risk. It shows its reasoning step by step and keeps it in the conversation, every answer leads to its source, feedback reaches whoever manages the agent and, from the Company plan, the audit log records activity.

  7. Art. 15

    Accuracy, robustness and cybersecurityVoluntaryhigh-riskOfficial text

    What it requires

    For high-risk systems: an appropriate level of accuracy, robustness and cybersecurity throughout the lifecycle.

    What Neurally does

    Neurally operates an information security management system certified to ISO/IEC 27001, and every answer can be checked against the passage it comes from.

  8. Art. 50.1

    Transparency: interacting with AINeurallyfrom 2 Aug 2026Official text

    What it requires

    People who interact with an AI system must be told so, unless it is obvious.

    What Neurally does

    In Neurally you talk to an AI agent that says it is one, and answers carry the notice “AI can make mistakes. Check important information.”

  9. Art. 50.2

    Marking of generated contentNeurallyby 2 Dec 2026Official text

    What it requires

    Generated text, images, audio and video must be marked in a machine-readable format and be detectable as AI-generated. For systems already on the market, by 2 December 2026.

    What Neurally does

    Neurally marks the documents it generates.

  10. Art. 50.4

    Text published to inform the publicYour companyfrom 2 Aug 2026Official text

    What it requires

    Anyone who publishes AI-generated text to inform the public on matters of public interest must disclose it, unless it has been reviewed by a person with editorial responsibility.

    What Neurally does

    Every document Neurally produces starts as a draft to be checked. If you publish generated text to inform the public, have a person review it or disclose it.

  11. Art. 53–55

    General-purpose AI modelsModel providersfrom 2 Aug 2025Official text

    What it requires

    Model documentation, information for those who integrate the models, copyright compliance, a summary of training data; for models with systemic risk, testing and mitigation.

    What Neurally does

    These obligations fall on model developers. Neurally integrates third-party models without retraining them and is responsible for the system’s obligations.

Four things to do, as a deployer.

Neurally does its part as a provider. These, on the other hand, are the responsibilities of the people who use it.

  1. 01

    Train your people

    Explain to the people who use Neurally how it works and what its limits are, and keep a record of the training (Art. 4).
  2. 02

    Stay clear of the prohibitions

    Don’t use AI for the practices prohibited by Art. 5, such as emotion recognition in the workplace.
  3. 03

    Assess high-risk uses

    If you plan to use it to select staff, assess creditworthiness or in other Annex III areas, assess this first (Art. 6).
  4. 04

    Review what you publish

    Generated text published to inform the public must be reviewed by a person or disclosed (Art. 50.4).

Penalties (Art. 99)

The higher amount applies; for SMEs and start-ups, the lower one.
  • €35mor 7% of worldwide turnover for prohibited practices.
  • €15mor 3% for other obligations, including Articles 26 and 50.
  • €7.5mor 1% for incorrect, incomplete or misleading information given to authorities.

Read the rules at the source.

Like Neurally’s answers, this page cites its sources: the text of the Regulation and the documents of the European Commission.

Learn more about data security

An AI that follows the rules. And proves it with sources.

Try Neurally on your documents, or ask us how it fits into your company’s processes.

Start your free trial7-day free trialNo credit card required

Request a demo.

We’ll show you Neurally at work on your team’s cases.

All fields are required. By submitting this form you accept the privacy policy (in Italian).