EU AI Act: What does the new AI legislation mean for customer service?
With the EU AI Act, the European Union has adopted the world’s first comprehensive regulation for artificial intelligence. From 2 August 2026, central transparency obligations will apply that are also relevant to companies using AI in customer service. What must companies disclose when using chatbots and AI agents? Do AI-generated replies need to be labelled? And what rules apply to emotion recognition and human oversight? This article explains the most important requirements and what they mean for customer service.
Why does the EU AI Act exist?
The EU AI Act is intended to ensure that artificial intelligence is used responsibly. It takes a risk-based approach: the applicable requirements depend on the purpose and specific use of an AI system. Different rules apply to prohibited practices, high-risk systems and applications that pose particular transparency risks. Standard chatbots and AI agents used in customer service are generally not classified as high-risk systems solely because they are used in support. For these systems, the transparency obligations under Article 50 are particularly relevant.
What rules apply to AI in customer service?
Companies that use AI-powered systems such as chatbots, AI agents or automated customer service processes must pay particular attention to the transparency obligations that apply from 2 August 2026:
Get practical guidance on AI transparency
1. Transparency obligation: customers must know when they are interacting with AI
Providers of AI systems that interact directly with people must ensure that the individuals concerned are informed about the use of AI. This information must be provided clearly, understandably and accessibly no later than at the beginning of the first interaction. Additional labelling is only unnecessary when it is obvious from the circumstances and context of use that the system is AI. Companies should therefore review their chatbots and virtual assistants and clearly identify them as AI.
2. Not every AI-generated reply requires visible labelling
The EU AI Act does not require every customer service email or chat response created with AI to be visibly labelled as “AI-generated”. Providers of generative AI systems must generally mark AI-generated or manipulated content in a way that makes it detectable as such in a machine-readable format.
Companies using AI must disclose, in particular, deepfakes and AI-generated publications on matters of public interest. For such texts, the disclosure obligation does not apply if they have undergone human review or editorial control and a natural or legal person holds editorial responsibility. Individual customer service replies do not automatically fall under this visible disclosure obligation.
3. Specific rules apply to emotion recognition
The EU AI Act significantly restricts certain forms of emotion recognition. Inferring emotions in the workplace from biometric data is generally prohibited. Limited exceptions apply for medical or safety-related purposes. Where emotion recognition or biometric categorisation systems are permitted, the individuals exposed to them must be informed about their use.
This must be distinguished from analysing written messages, for example to identify positive or negative sentiment in a support ticket. Text-based sentiment analysis is not automatically considered biometric emotion recognition under the EU AI Act. The data processed and the specific way in which the system works are decisive.
4. Human handovers must be defined appropriately
The EU AI Act does not generally require standard customer service chatbots to allow customers to switch to a human agent at any time. Companies should nevertheless define the concerns, uncertainties and risks that require an AI agent to hand the request over to a human.
This may include complaints, legally relevant declarations, particularly sensitive cases or situations in which the AI agent cannot resolve a request reliably. Depending on the use case, requirements for human oversight may also arise from other laws or contractual obligations.
What does this mean for companies that use Zendesk?
Companies using Zendesk AI Agents, automated replies or chatbots should first document which systems interact directly with customers and which tasks they perform. They should then review the applicable transparency and information obligations for each channel. This includes:
- Labelling AI agents and chatbots: Inform customers clearly, no later than at the beginning of the first interaction, that they are communicating with an AI system.
- Reviewing each channel separately: Check the labelling and presentation in web, mobile and social messaging as well as other channels.
- Distinguishing between types of AI-generated content: Determine whether the content requires machine-readable marking or visible disclosure.
- Distinguishing emotion recognition from sentiment analysis: Clarify whether the system uses biometric emotion recognition or only analyses sentiment in written text.
- Defining handover rules: Determine when an AI agent should stop processing a request and hand it over to a human.
- Documenting responsibilities: Clarify which obligations apply to Zendesk as the provider and which apply to the company using the system.
- Ensuring AI literacy: Provide employees who use AI systems or assess their results with the necessary knowledge and skills.
What deadlines apply?
The EU AI Act is becoming applicable in stages. The following dates are particularly relevant to companies using AI in customer service:
- 2 February 2025: The prohibitions on certain AI practices and the obligation to ensure sufficient AI literacy became applicable. Companies must ensure that employees who use or oversee AI systems have an appropriate level of knowledge.
- 2 August 2025: The first requirements for providers of general-purpose AI models became applicable. Longer transitional periods apply to some models that were already available before this date.
- 2 August 2026: The transparency obligations under Article 50 apply. These include information about direct interactions with AI systems and requirements for the machine-readable marking and disclosure of certain AI-generated or manipulated content.
- 2 August 2027: Further requirements apply to certain AI systems that are used as safety components of regulated products or are themselves regulated products.
The relevant deadline depends on the company’s role, the system being used and its intended purpose. A company may simultaneously be a provider, a deployer or part of a wider AI value chain.
What are the penalties for non-compliance?
The potential fines depend on the type of infringement. Violations involving prohibited AI practices may result in fines of up to €35 million or 7 % of the company’s total worldwide annual turnover for the preceding financial year. Other infringements of obligations under the EU AI Act may generally result in fines of up to €15 million or 3 % of total worldwide annual turnover. The specific penalty depends on factors including the nature, severity and duration of the infringement as well as the size of the company.
Conclusion: review your transparency obligations now
Review your AI systems and processes.
From 2 August 2026, central transparency obligations under the EU AI Act apply. Companies using AI in customer service should document which systems interact directly with customers, how these systems are identified and which content is generated fully or partially by AI.
The exact function of the system matters: an AI agent that communicates with customers independently is subject to different requirements than a Copilot that supports human agents. Text-based sentiment analysis, biometric emotion recognition and AI-generated publications must also be distinguished from one another.
Leafworks supports you in reviewing your Zendesk configuration and AI-powered processes – from labelling AI agents and designing bot flows and handover rules to technical and organisational documentation.
Sources and further information
- Official text of the EU AI Act: Regulation (EU) 2024/1689
- European Commission – Article 50: Transparency obligations for providers and deployers of certain AI systems
- European Commission guidelines: Transparency obligations for AI-generated content
- European Commission questions and answers: Transparency obligations under Article 50
- Zendesk Trust Center: Security and privacy at Zendesk
