Brief Statement on the EU AI Act for LoyJoy

Last reviewed on

Change history (9 entries)
  • 07/09/2026Standardised LoyJoy's role as provider. Separated the transparency notice under Art. 50(1) from the machine-readable marking of synthetic outputs under Art. 50(2). Expressly excluded high-risk applications in line with the Acceptable Use Policy.
  • 18/08/2026Added general guidance on risk classification under the EU AI Act.
  • 10/08/2026The speech-to-speech model of the Phone Agent now runs in the EU Data Zone of Azure OpenAI instead of via global Azure services. Data Protection section updated accordingly.
  • 05/08/2026Added Digital Omnibus update: high-risk deadlines postponed to December 2027 and August 2028, Art. 50 transparency obligations unaffected and unchanged in force since August 2, 2026. Added December 2, 2026 deadline for existing systems regarding labelling of synthetic audio content (technical implementation under review). Added deadline timeline and FAQ section with FAQPage schema.
  • 30/07/2026Split the Transparency Obligations section into 'Chat' and 'Phone Agent'. Added chat transparency details: AI notice in the greeting module, customizable AI label on AI-generated messages, permanent notice line below the input field. Made the labelling of AI chat messages a binding statement.
  • 03/07/2026Clarification added that LoyJoy operates the GPAI model Gemma 4 on its own hardware in the Münster data center. Role under Art. 3 no. 3 and Art. 25(1)(c) AI Act clarified.
  • 06/05/2026Standard Phone Agent greeting documented. Platform note on responsibility for modified greetings added.
  • 05/05/2026LoyJoy role extended to include Phone Agent and speech-to-speech. Transparency obligations for Phone Agent conversations added. Obligations table updated.
  • 20/05/2025Initial publication of the AI Act statement.

Deadlines at a glance

  1. August 2, 2026

    The transparency obligations under Art. 50 AI Act take effect, for direct contact with AI systems via chat and phone.

  2. December 2, 2026

    Implementation deadline for machine-readable marking of synthetic audio, image, video and text outputs from generative systems already on the market before August 2, 2026.

  3. December 2027

    New deadline under the Digital Omnibus for part of the high-risk obligations.

  4. August 2028

    New deadline under the Digital Omnibus for further high-risk obligations.

Brief Statement on the EU AI Act for the LoyJoy Platform

This brief statement explains how LoyJoy, as a platform provider, implements the obligations of the EU AI Act, in particular the transparency obligations under Art. 50 for direct contact with AI systems via chat and phone.

Digital Omnibus

What was postponed, and what wasn't

Under the "Digital Omnibus", the deadlines for high-risk AI systems were postponed, to December 2027 and August 2028 respectively.

Not postponed are the transparency obligations under Art. 50 AI Act for direct contact with AI systems. These have applied unchanged since August 2, 2026.

Role of LoyJoy

LoyJoy provides enterprise customers with a SaaS platform for creating and running AI-powered customer interactions via chat and telephone. LoyJoy does not develop its own foundation models.

As the default model for chat interactions, LoyJoy operates the GPAI model Gemma 4 from Google on LoyJoy-owned hardware in the Münster data center. Google remains the provider of the GPAI model within the meaning of Art. 3 no. 3 and Art. 51 et seq. AI Act; the GPAI-model provider obligations continue to rest with Google. LoyJoy does not carry out any substantial modification of the model within the meaning of Art. 25(1)(c) AI Act (no fine-tuning, no further training), and the on-prem hosting alone does not turn LoyJoy into a provider of the GPAI model. LoyJoy continues to act as the provider of the AI system built on top of the model.

In addition, depending on the configuration, AI models and speech-to-speech services from sub-processors such as Microsoft Azure, Mistral, Scaleway, Nebius or Anthropic are used.

In relation to end users, the customer is generally the deployer of the specific AI system and the controller responsible for the specific use case. LoyJoy supports customers as a platform provider and data processor through technical and organisational features for transparent and GDPR-compliant use.

Risk Classification under the EU AI Act

The LoyJoy Platform is not assigned to a specific risk class solely because of the language model it uses. The decisive factor is the specific intended purpose. The contractually intended purpose of the LoyJoy Platform is limited to AI-powered customer interactions via chat and phone outside high-risk applications.

Pure information, communication and support functions are generally not classified as high-risk applications as long as the AI system does not make decisions about individuals or materially influence such decisions. Typical examples include answering questions from approved knowledge sources, providing general product and service information, navigating processes, and capturing and forwarding requests.

LoyJoy is not intended for use as a high-risk AI system within the meaning of Article 6 in conjunction with Annex I or III of the EU AI Act. Such use is excluded under the Acceptable Use Policy. This includes certain applications in employment, education, creditworthiness assessment, essential public and private services, and risk assessment and pricing in life and health insurance.

The customer defines the specific intended purpose within the contractually permitted use and assesses it before deployment and following material changes. LoyJoy supports the assessment and secure use through configurable guardrails, process constraints, logging, human handover and technical documentation. Further guidance on classification is available from the European Commission.

Transparency Obligations

The transparency obligations under Art. 50 AI Act have applied since August 2, 2026, and are not affected by the postponement of the high-risk deadlines under the Digital Omnibus.

Chat

For chat interactions, LoyJoy provides an AI notice in the greeting module that informs the user at the start of the conversation that they are interacting with an AI system. Every AI-generated message is additionally labelled with a customizable AI label. LoyJoy also displays a permanent notice line below the input field, visible throughout the entire conversation.

Ready-to-use text templates for the greeting module and notice line are available under AI Notice Text Templates.

Phone Agent

For Phone Agent conversations, LoyJoy provides a configurable default greeting that informs the caller at the beginning that they are interacting with an AI system. The default greeting reads:

“Hello, I am the AI telephone assistant of [Company name]. The call will be temporarily stored to process your request, and of course treated confidentially. How can I help you?”

Customers may adapt this greeting to their use case. The platform includes the following notice: “The greeting should clearly state that the caller is speaking with an AI system. If you remove this notice, you are responsible for the legal assessment of your modified greeting.”

If call recordings are used, the controller must additionally assess whether and in what form a separate notice or consent from callers is required for the recording.

Ready-to-use text templates for the phone greeting are also available under AI Notice Text Templates.

Obligations and Implementation

ObligationImplementation at LoyJoy
Transparency for direct AI interaction, Art. 50(1) AI ActAI-generated chat messages are labelled through an AI notice in the greeting module, a customizable AI label, and a permanent notice line. For Phone Agent interactions, LoyJoy provides a default greeting with a clear AI notice.
Machine-readable marking of synthetic outputs, Art. 50(2) AI ActThis technical marking obligation is separate from the human-perceptible AI notice under Art. 50(1). For generative systems already on the market before August 2, 2026, an implementation deadline of December 2, 2026 applies. The technical implementation and applicability to the output types generated by LoyJoy are currently under review.
LoggingConversation data, including transcripts and call recordings, is stored according to tenant configuration, 30 days by default.
Human oversightCustomers can configure and review process logic, knowledge sources and responses. A handover to human agents can be included per process.
CybersecurityEncrypted transmission, role-based access control, logging, sub-processors with documented security measures.

Data Protection

  • Processing as a data processor pursuant to Art. 28 GDPR.
  • By default, Gemma 4 is operated as the chat LLM on LoyJoy-owned hardware in the Münster data center (Germany). For this default path, no personal data is transferred to an external AI model provider.
  • Further LLM processing takes place in EU regions by default. LoyJoy offers a selection of AI models from various providers operated in the EU. For customers not subject to the EU AI Act, LoyJoy additionally makes AI models with server locations in the USA available. These are clearly labelled in the platform and are not configured as the default for EU customers.
  • Since August 2026, the speech-to-speech model of the Phone Agent has been processed in the EU Data Zone of Azure OpenAI.
  • Encrypted storage and transmission of personal data.

Frequently Asked Questions