The AI Voice That Answers Your Call: What Europe Now Expects You to Disclose

Written by Safegram Editorial Team

An AI receptionist can take an enquiry while a small business is busy. A virtual creator can explain a product without a camera crew. But when a realistic voice, face or message reaches a customer, the trust question is straightforward: do they know they are dealing with AI?

The EU’s AI Act transparency requirements under Article 50 began applying on 2 August 2026. A second date, 2 December 2026, is approaching, but it is not the start of every AI-label rule. The later deadline is a narrow transition for providers of generative systems already on the market before 2 August to implement the machine-readable marking and detection requirement. The European Commission explains the distinction in its official FAQ. [1][3]

The important difference

• AI interaction: a provider of a directly interactive system must design it to make people aware of the AI interaction, unless the AI nature is obvious. [1]
• Content marking: providers of systems generating synthetic audio, images, video or text have obligations for detectable, machine-readable marking, subject to scope and exemptions. [1][5]
• Human-facing disclosure: businesses and professional users who deploy deepfakes must disclose them visibly or audibly; this is separate from technical marking. [1]

Why the December deadline is often misunderstood

The Council adopted simplifying AI legislation on 29 June 2026 and described the transition for older generative systems as ending on 2 December. The Commission’s guidance, published 20 July, explains that Article 50’s broad transparency obligations applied from 2 August. The December transition applies only to particular provider-side marking and detection obligations for systems placed on the market before that August date. [2][3]

A company launching a new generator after 2 August should not read “December” as permission to ignore the rules until then. Equally, a small company using an AI assistant should not assume it owns every technical duty of the model developer. Legal roles matter: the provider develops or markets the AI system under its name; the deployer uses it professionally under its authority. Sometimes a business can occupy both roles. [1]

A creator examining footage on a laptop before a commercial post; AI-generated editorial illustration.

A useful assistant should introduce itself

Imagine a fictional Dublin beauty salon using an AI phone receptionist to take requests outside opening hours. It may save staff time and help a caller leave details. It should not make a caller believe a human employee is on the line if the interaction is with AI. The Commission says notification should be clear from the start of the first direct interaction, with a narrow exception when the AI nature is obvious. [1]

The practical design question is not whether a tiny footnote exists somewhere in the company’s terms. It is what the caller actually hears or reads. A clear introduction such as “You are speaking with an automated AI assistant” is an illustrative design suggestion, not a legal safe-harbour formula. Businesses should test how people understand the interaction and how they can reach a human when appropriate.

An AI creator, an altered image and a deepfake are not identical

Article 50 does not say every photograph edited with software must carry the same warning. The Commission defines deepfakes by resemblance to real or plausible people, places, objects or events and their potential to appear authentic when they are not. The context matters: a clearly imaginary illustration and a fabricated video impersonating a real local business owner raise very different concerns. [1]

A professional deployer of a deepfake has a disclosure obligation to the people exposed to it. A hidden technical mark alone does not satisfy that human-facing duty. Artistic or satirical works have a tailored form of disclosure, and standard editing may be outside the separate technical-marking rule. Creators should record what was generated, what was filmed and what substantive changes were made before deciding how to present it. [1][5]

Two customers checking a phone with an abstract voice waveform; AI-generated editorial illustration.

A paid partnership needs another kind of honesty

Irish consumer law adds a separate question: was there a commercial benefit or arrangement behind the recommendation? The CCPC’s influencer guidance requires commercial content to be recognisable as advertising in relevant circumstances; correct ad labelling does not excuse a misleading claim. An AI-content disclosure cannot substitute for an applicable advertising disclosure, and #Ad cannot by itself explain a deceptively realistic synthetic speaker. [4]

Consider a hypothetical creator demonstrating a local service with a synthetic avatar. The creator should consider whether the presentation needs an AI disclosure, clearly identify any commercial arrangement, and avoid implying that a fictional avatar personally purchased or tested the service. This is a proposed good practice, not an allegation about a named creator or an account on Safegram.

Public-interest writing and genuine editorial control

The EU rules also address certain AI-generated text published to inform the public on matters of public interest. The Commission describes an exemption where there has been meaningful human review or editorial control and a person holds editorial responsibility. A superficial spelling check is not substantive review. That distinction matters for news publishers as well as businesses publishing guides about law, consumer safety or financial decisions. [1][6]

For Safegram News, the responsible approach is to disclose that translation and drafting can be machine-assisted, retain named primary sources and seek substantive human editorial review before publishing. An automated check must not be represented as independent human sign-off. The aim is clarity, not a claim that every synthetic sentence is deceptive.

A small production team reviewing a fictional virtual presenter on a monitor; AI-generated editorial illustration.

Four steps a small business can take

First, inventory where AI speaks directly to people, where it generates media and where a human or company distributes that media. Second, distinguish the provider’s technical marking duty from the deployer’s content disclosure. Third, check commercial advertising and consumer-protection requirements separately. Fourth, keep an approval record showing who checked claims and whether any realistic voice or image could mislead. [1][4][5]

The Commission’s transparency code is voluntary: it can help signatories demonstrate compliance, but not signing does not remove statutory duties. Nor does a technical watermark prove a review, testimonial or booking is genuine. Trust needs multiple signals. [1][5]

Safegram’s emphasis on real people and user control fits a wider lesson: identity signals, truthful advertising and clear AI disclosures answer distinct questions. A verified business is not proof that every message was human-written, and an AI label is not proof that an offer is accurate. Explore the platform at safegram.com, while applying those checks across every service you use. This article is general editorial analysis, not legal advice or a certification of any particular implementation.

Frequently asked questions

Did EU AI transparency law start on 2 December 2026?

No. Article 50 applies from 2 August 2026. The 2 December date is a limited transition for machine-readable content marking by generative systems placed on the market before 2 August. [1][3]

Must an AI receptionist tell callers it is AI?

Providers of systems that directly interact with people must design them to make the AI interaction clear unless it is obvious. The disclosure should be apparent at the beginning of the interaction; particular implementation and roles matter. [1]

Is every AI-assisted photograph a deepfake?

Not automatically. The definition and context matter, including resemblance to a real or plausible subject and the risk of falsely appearing authentic. Standard editing may fall outside the relevant marking rule. [1]

Does a machine-readable mark replace an on-screen disclosure?

No. For deepfakes, the deployer’s human-perceptible disclosure is a separate obligation; provider-side hidden marking is not enough. [1]

Can an influencer use AI and simply add #Ad?

An advertising disclosure and an applicable AI disclosure solve different problems. Neither guarantees an accurate product claim or an authentic endorsement. Check the obligations for the actual content and commercial arrangement. [1][4]

Sources

[1] European Commission — Article 50 transparency FAQ, checked 5 November 2026 https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act

[2] European Commission — Article 50 guidelines, published 20 July 2026 https://digital-strategy.ec.europa.eu/en/news/commission-publishes-guidelines-transparency-obligations-providers-and-deployers-certain-ai-systems

[3] Council of the EU — AI rules simplification adopted 29 June 2026 https://www.consilium.europa.eu/en/press/press-releases/2026/06/29/artificial-intelligence-council-gives-final-green-light-to-simplify-and-streamline-rules/

[4] Irish CCPC — Influencer advertising and marketing guidance, checked 5 November 2026 https://www.ccpc.ie/information-for-businesses/guidance-for-businesses/consumer-protection-guidance/influencer-advertising-and-marketing

[5] European Commission — Code of Practice on AI-generated-content transparency https://digital-strategy.ec.europa.eu/en/faqs/code-practice-transparency-ai-generated-content

[6] EU AI Act Service Desk — Article 50 consolidated legal text https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50

[7] European Commission — Guidelines policy overview, updated 6 August 2026 https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations