EU AI Act and GDPR for business AI

Since 2 August 2026, anyone talking to an AI voice agent in the EU must be told it is an AI at the latest at the first interaction (AI Act Art. 50). High-risk duties apply from 2 December 2027, and only to the uses listed in Annex III, such as credit scoring. Calls, recordings and documents are personal data, so the GDPR applies in full, wherever the AI runs.

What applies when

Application dates of the EU AI Act after the Digital Omnibus on AI (Art. 113 as amended)
DateWhat appliesWhat it means for business AI
2 February 2025Definitions, AI literacy, prohibited practices (Art. 5)No manipulative techniques, no inferring the emotions of employees at work
2 August 2025Penalties (Art. 99), general-purpose AI rulesFines can be imposed for prohibited practices
2 August 2026General application, including transparency (Art. 50)People must be told they are talking to an AI, from the first interaction
2 December 2026Machine-readable marking of synthetic audio for systems already on the market (Art. 111(4))A provider duty, not one for the business using the agent
2 December 2027High-risk duties for Annex III usesOnly for listed uses such as credit scoring or life and health insurance pricing
2 August 2028High-risk duties for AI in regulated products (Annex I)Machines, medical devices and similar products

Sources: EU AI Act (Regulation (EU) 2024/1689), Art. 113; Digital Omnibus on AI (Regulation (EU) 2026/1744); EU AI Act (Regulation (EU) 2024/1689), Art. 111(4)

Provider or deployer: who does what

The AI Act splits duties between the provider, who develops an AI system and places it on the market under its own name (Art. 3(3)), and the deployer, who uses it under its own authority (Art. 3(4)). For the AI voice agents, AI Fortis is the provider; the company whose customers the agent talks to is the deployer.

The provider must design an agent that talks to people so that they are informed they are dealing with an AI (Art. 50(1)) and must mark synthetic audio in a machine-readable way (Art. 50(2)). The deployer has its own duties if it uses emotion recognition (Art. 50(3)) or publishes deep fakes (Art. 50(4)), and the deployer duties of Art. 26 if its use is high-risk. The information must be given at the latest at the first interaction (Art. 50(5)).

Sources: EU AI Act (Regulation (EU) 2024/1689), Art. 3(3); EU AI Act (Regulation (EU) 2024/1689), Art. 3(4); EU AI Act (Regulation (EU) 2024/1689), Art. 50(1); EU AI Act (Regulation (EU) 2024/1689), Art. 50(2); EU AI Act (Regulation (EU) 2024/1689), Art. 50(5); EU AI Act (Regulation (EU) 2024/1689), Art. 26

Is your use of AI high-risk?

Typical uses of our products and whether the EU AI Act lists them in Annex III
UseListed in Annex III?Note
Customer hotline, bookings, reminders, surveysNoTransparency (Art. 50) and the GDPR apply
Deciding a person's creditworthiness or credit scoreYes, point 5(b)Fraud detection is excluded from this point
Risk assessment and pricing in life and health insuranceYes, point 5(c)Claims calls and policy questions are not this decision
Classifying emergency calls or triaging patientsYes, point 5(d)Appointment booking is not triage
Scoring the performance of human call-centre agentsYes, point 4(b)Inferring employees' emotions is prohibited (Art. 5(1)(f))
Invoice extraction, archiving, ERP, VAT journalsNoAI literacy (Art. 4) and the GDPR apply

Sources: EU AI Act (Regulation (EU) 2024/1689), Art. 6(2); EU AI Act (Regulation (EU) 2024/1689), Annex III, point 4(b); EU AI Act (Regulation (EU) 2024/1689), Annex III, point 5(b); EU AI Act (Regulation (EU) 2024/1689), Annex III, point 5(c); EU AI Act (Regulation (EU) 2024/1689), Annex III, point 5(d); EU AI Act (Regulation (EU) 2024/1689), Art. 5(1)(f)

The GDPR for calls and documents

Call audio, transcripts, a caller's voice and the data on an invoice relate to identifiable people, so they are personal data (GDPR Art. 4(1)). The company that decides why and how they are processed is the controller; a supplier that processes them on its behalf is a processor (Art. 4(7) and 4(8)) and needs a contract with the content of Art. 28(3).

Callers must get the information of Art. 13 when the data is collected, which in practice means a short notice at the start of the call and a link to the full text. Health data from patient calls is a special category (Art. 9), and a voice used to identify a person is biometric data (Art. 4(14)). New technology at scale usually calls for a data protection impact assessment (Art. 35), and security measures such as encryption are required by Art. 32.

Where the AI runs changes who processes the data, not whether the GDPR applies. On-premises, the data stays with the controller; in the cloud, the provider is a processor bound by the data processing agreement.

Sources: GDPR (Regulation (EU) 2016/679), Art. 4(1); GDPR (Regulation (EU) 2016/679), Art. 4(7); GDPR (Regulation (EU) 2016/679), Art. 4(8); GDPR (Regulation (EU) 2016/679), Art. 28(3); GDPR (Regulation (EU) 2016/679), Art. 13; GDPR (Regulation (EU) 2016/679), Art. 9; GDPR (Regulation (EU) 2016/679), Art. 35; GDPR (Regulation (EU) 2016/679), Art. 32

Outbound calls and call recording

Marketing calls made by an automated calling system without human intervention need the person's prior consent in every Member State (ePrivacy Directive Art. 13(1)); an AI voice agent making sales calls falls under this rule. Service calls a customer asked for, and inbound calls, are a different case.

Recording calls is governed by national law implementing Art. 5 of the ePrivacy Directive, which allows legally authorised recording in the course of lawful business practice as evidence of a commercial transaction (Art. 5(2)). National rules differ, so check yours before switching recording on.

Sources: ePrivacy Directive (Directive 2002/58/EC), Art. 13(1); ePrivacy Directive (Directive 2002/58/EC), Art. 5(2)

Claims to be careful with

  • There is no general "AI Act certificate". Conformity assessment and CE marking apply to high-risk systems only.
  • The transparency duty of Art. 50 was not postponed: it applies from 2 August 2026.
  • Running AI on your own servers does not switch off the GDPR, DORA or NIS2. It changes who processes the data and where.
  • The separate proposal to amend the GDPR and the cookie rules is not adopted; the current texts apply.
  • A voice agent that answers questions is not a high-risk system just because it uses AI. The use decides.

Frequently asked questions

Does the EU AI Act apply to AI voice agents?

Yes. Since 2 August 2026, people must be told they are talking to an AI at the latest at the first interaction (Art. 50(1) and 50(5)). Most customer-service uses are not high-risk.

When do the high-risk rules apply?

From 2 December 2027 for the uses listed in Annex III and from 2 August 2028 for AI in regulated products, after the Digital Omnibus on AI (Regulation (EU) 2026/1744) moved the dates.

Is a voice bot high-risk under the AI Act?

Not by itself. It becomes high-risk when it is used for a listed purpose, for example to decide creditworthiness (Annex III, point 5(b)) or to classify emergency calls (point 5(d)).

Who has to tell callers they are talking to an AI?

The provider must build the agent so that people are informed (Art. 50(1)). In our agents, the AI disclosure at the start of the call is on by default.

Is there an AI Act certification for AI products?

No general one. Conformity assessment and CE marking exist only for high-risk systems. Be careful with any product sold as "AI Act certified".

Does on-premises AI make GDPR compliance easier?

It keeps call audio and documents inside your own infrastructure, so no external processor is in the path. The GDPR still applies in full to your own processing.

Talk to us about your case

Tell us which AI system you plan to use and where your data must stay; we show you how our products handle the rules. We reply personally, usually within one business day.