EU AI Act for small businesses: what really applies to you
For a business using an AI phone assistant, little of the EU AI Act remains in practice: the duty to tell callers an AI is speaking (Article 50(1), since 2 August 2026) falls on the provider. As the deployer, you support your team's AI literacy (Article 4). The heavy duties apply to high-risk systems; call answering is not among them.
This article is orientation, not legal advice. For your individual case the topic belongs with your legal or data protection advisers.
An all-clear with substance — not with a shrug
Two equally wrong stories circulate about the EU AI Act: "only affects large corporations, ignore it" and "a bureaucratic monster that crushes every business". Neither is true. The AI Act (Regulation (EU) 2024/1689) is tiered by risk. Certain practices are prohibited, manipulative techniques for instance (Article 5). High-risk systems carry the big package of duties, for instance AI that filters job applications or assesses people's creditworthiness (Annex III, points 4 and 5). Systems that interact directly with people are subject to a transparency duty (Article 50(1)) — that is where an AI phone assistant belongs. And for all AI systems, providers and deployers must take measures that support AI literacy (Article 4).
This article reflects the law as of 15 November 2026. It takes in the amendments made by the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026) and the European Commission's guidelines on Article 50 of 5 November 2026.
Provider and deployer duties: who shoulders what
The AI Act distinguishes the provider of an AI system — whoever develops it or has it developed and places it on the market or puts it into service under their own name (Article 3(3)) — from the deployer, who uses it under their own authority (Article 3(4)). If you use an AI phone assistant, you are the deployer; whoever builds the assistant and offers it under their own name is the provider. At scoco.ai this allocation of roles is also written into the terms and conditions (AGB § 7(2)).
As long as the assistant is not a high-risk system, it follows that:
- The provider bears the disclosure duty: it has to design the system so that callers learn they are speaking to an AI (Article 50(1)).
- Provider and deployer are both bound by Article 4: measures that support the AI literacy of the people working with the system on their behalf.
- Not for you are the deployer duties in Article 26, such as use in line with the instructions for use, human oversight and keeping logs. They apply to high-risk systems only.
The Digital Omnibus softened Article 4. Until 26 July 2026 it required measures to ensure, "to their best extent", "a sufficient level of AI literacy" among staff. Since 27 July 2026 it requires measures that support the development of AI literacy, expressly without any duty to guarantee a specific level for any individual. According to the European Commission's questions and answers, no certificate is needed for this, and no particular governance structure such as an "AI officer" either. For documentation the Commission names an internal record of trainings and guidance. How much is appropriate depends on your team and the use; simply asking staff to read the instructions for use may, according to the Commission, be ineffective in many cases.
Article 50: the provider's duty — and your part in it
Article 50 has applied since 2 August 2026 (Article 113). The Digital Omnibus did not postpone the disclosure duty in paragraph 1. It deferred only the machine-readable marking of AI-generated content under paragraph 2, for systems already on the market before that date, to 2 December 2026 (Article 111(4); guidelines para 153). Paragraph 1 is addressed to the provider: it has to design and develop the system so that people are informed they are interacting with an AI system — unless that is obvious to a reasonably well-informed, observant and circumspect person. The information has to come at the latest at the first interaction, in a clear and distinguishable manner (paragraph 5).
What that means on the phone is set out in the Commission's guidelines (C(2026) 7702, non-binding). There is no prescribed wording (para 34). For phone and voice they give as an example an explicit spoken statement at the start, such as "This is an AI-powered assistant" (para 37). In their view a single prominent notice before the first interaction is likely to suffice in most cases. In riskier situations, such as health, legal or financial questions, complaints or vulnerable callers, they consider periodic reminders likely to be necessary; and if someone asks, the system has to say that it is an AI (para 40). Ambiguous signals such as a generic reference to an "assistant", and notices that sit only in terms and conditions or documentation, are not meant to be enough (para 38).
Your part as the deployer is practical: choose a provider that builds the disclosure in, and do not undermine it. Where you write the greeting yourself, you help decide how clear the notice is. At scoco.ai you can reword the AI disclosure in the greeting, but you cannot remove it. scoco checks when you save that an accepted phrase ends within the first 80 characters, and before every call that it is there; the terms and conditions forbid circumventing the notice (AGB § 7(3)). The check also accepts phrases without the word "KI" (AI), such as „digitaler Assistent" (digital assistant) or „Sprachassistent" (voice assistant) (the full list is in the help centre). With para 38 in mind, a phrase that names „KI" or „künstliche Intelligenz" (artificial intelligence) explicitly is the clearer choice, for example „Hier ist die KI-Assistentin von …" (this is the AI assistant at …). More on this in the article "Must the AI say it is an AI?".
What you should actually do as a business
The list for the typical case is short:
- Check the greeting: call once yourself. Does the greeting say explicitly, right at the start, that an AI is speaking?
- Ask the provider: is the disclosure built in for good, or can it be switched off? Does the assistant repeat it when it switches language? Get the answers in writing.
- Brief the team (Article 4): anyone working with transcripts and summaries should know what the system can and cannot do — for instance, that AI output can be wrong. Keep a note of who was briefed and when.
- Handle the GDPR in parallel: the AI Act replaces no data protection duty; according to the guidelines, Article 50 does not affect the information duties of data protection law (para 52). The data processing agreement (Article 28 GDPR), informing callers (Article 13 GDPR) and retention periods all carry on alongside it. Informing callers remains your job; non-binding wording help for your privacy notice is in the help centre under "Informing callers".
That is manageable, not a major project.
Honestly: where you have to look more closely
Three situations deserve more than this article. First: anyone who uses AI beyond call answering, for selecting applicants or assessing creditworthiness, say, can fall into the high-risk category. For the areas in Annex III those duties apply from 2 December 2027 under the Digital Omnibus (Article 113). The assistant itself also needs checking if it does more than take down requests: Annex III lists, among other things, systems that evaluate emergency calls and emergency healthcare patient triage systems (point 5(d)). Whether an assistant that passes calls to people by urgency can be caught by that is not settled — clarify it for your case with your legal adviser. At scoco.ai, use in high-risk cases is permitted under the terms and conditions only after prior agreement (AGB § 7(6)).
Second: anyone who has an AI system developed and places it on the market or puts it into service under their own name or trademark is a provider themselves (Article 3(3)). Anyone who modifies an existing system and puts it into service under their own name becomes, according to the guidelines, the provider of the new system (para 11). That is remote for the ordinary user of an AI phone assistant, not remote for resellers.
Third: supervisory and enforcement practice under the AI Act is young. The Commission's guidelines are non-binding; only the Court of Justice of the EU can give an authoritative interpretation of the regulation (para 5), and the Commission intends to revise the guidelines as practice develops (para 155). Anyone working close to a grey area should settle it with legal advice rather than with blog articles — including this one.
Frequently asked questions
Is an AI phone assistant a high-risk system?
Annex III of the AI Act lists the high-risk areas, among them recruitment, creditworthiness, the evaluation of emergency calls and emergency healthcare patient triage. A business's call answering — taking down requests, booking appointments, organising callbacks — is not on it. If an assistant is used for a purpose on that list, those duties apply, for Annex III from 2 December 2027; that needs a legal review in the individual case.
Do I need an "AI officer" or a certification as a small business?
No. According to the European Commission's questions and answers on Article 4, AI literacy requires neither a certificate nor a particular governance structure such as an AI officer; for documentation the Commission names an internal record of trainings. What is required are measures that support your team's AI literacy (Article 4 as amended by the Digital Omnibus). A conformity assessment concerns high-risk systems only (Article 43).
What happens if the AI disclosure is missing?
For infringements of the transparency duties in Article 50, Article 99(4)(g) provides for fines of up to €15 million or, for undertakings, up to 3 % of worldwide annual turnover, whichever is higher; for SMEs the lower of the two applies (Article 99(6)). The detailed penalty rules are set by the Member States (Article 99(1)). The addressee of the disclosure duty in Article 50(1) is the provider. Whether a missing disclosure can additionally be challenged under competition law (in Germany §§ 5, 5a UWG, misleading practices) depends on the individual case; according to the Commission's guidelines, consumer protection law applies alongside Article 50 (para 50).
Does the AI Act apply in addition to the GDPR or instead of it?
In addition. The AI Act governs the AI system, the GDPR the personal data; according to the Commission's guidelines, Article 50 does not affect the information duties of data protection law (para 52). For an AI phone assistant that means: the AI disclosure in the greeting under Article 50, and alongside it a data processing agreement (Article 28 GDPR), informing callers (Article 13 GDPR) and retention periods.
Status & sources
As of 2026-11-15.
- Not legal advice
- EU AI Act (Regulation (EU) 2024/1689): Art. 3(3) and (4), Art. 4, Art. 5, Art. 26, Art. 43, Art. 50(1) and (5), Art. 99(1), (4)(g) and (6), Art. 111(4), Art. 113, Annex III points 4 and 5
- Digital Omnibus on AI: Regulation (EU) 2026/1744 of 8 July 2026, OJ L of 24 July 2026, in force since 27 July 2026 (Art. 4 replaced; high-risk under Annex III from 2 December 2027; Art. 50(1) unchanged, transition only for Art. 50(2) to 2 December 2026)
- European Commission: Guidelines on the implementation of the transparency obligations under Article 50 of the AI Act, C(2026) 7702 final of 5 November 2026 (non-binding; paras 5, 11, 34, 37, 38, 40, 50, 52, 153, 155)
- European Commission: AI Literacy – Questions & Answers, last updated 20 October 2026 (no certificate, no mandated governance structure, internal record)
- GDPR (Regulation (EU) 2016/679) Articles 13 and 28
- §§ 5, 5a UWG (German Act against Unfair Competition: misleading practices)
- scoco.ai product information: as of 15 November 2026 (AGB § 7(2), (3) and (6); help page "Change the greeting")