Since 2 August 2026, the European AI Act imposes a simple rule on any system that talks to your customers: state clearly that it is an AI. At NoviaMind, that sentence has led every call from day one. We did not scramble into compliance: we already were compliant.
Here is what the text says, the timeline of obligations, and what it concretely changes for a voice agent in real estate.
The AI Act in one sentence
The AI Act (Regulation (EU) 2024/1689) is the first European framework that governs artificial intelligence according to its level of risk. It does not apply all at once: its obligations enter into force in stages between 2025 and 2027. A voice agent like NoviaMind's falls under the "limited risk" category, whose main obligation is transparency.
The timeline of obligations to remember
The text entered into force on 1 August 2024, then applies in stages:
- 2 February 2025: bans on unacceptable AI practices and the AI literacy obligation (Article 4). Teams that deploy and operate an AI must have a sufficient understanding of it, limitations included.
- 2 August 2025: obligations for general-purpose AI models and the setup of governance.
- 2 August 2026: the major deadline. The transparency obligations of Article 50 become applicable, along with a large part of the regime applicable to high-risk systems.
- 2 December 2026: grace period for machine-readable marking of generated content (Article 50(2)), a point still ambiguous for real-time voice.
- 2 August 2027: end of the transition period for certain high-risk systems embedded in already-regulated products.
The penalties are not symbolic: failing transparency obligations can cost up to 15 million euros or 3% of worldwide annual turnover.
Concretely, for a real estate voice agent
Article 50(1) directly targets systems that interact with people, such as a voice agent. The rule: the person must be informed they are talking to an AI, unless it is obvious. This is not a heavy constraint. It is a matter of posture: announce, or let them believe otherwise.
One point deserves emphasis, because it is a product boundary as much as a legal one. As long as the agent qualifies and hands off to a human who decides, it stays in "limited risk". If it decided in place of the human (for example rejecting a tenant application on its own based on a solvency criterion), it would shift to "high risk", with an entirely different level of obligations. At NoviaMind, that boundary is a design choice: the agent prepares, the human decides.
NoviaMind's position: transparency by design
The transparency obligation of Article 50, NoviaMind has met all along. From the first second of the call, the agent presents itself for what it is: the agency's voice assistant. No deception, no ambiguity, and above all: this is not a box ticked in a rush on 2 August 2026, it is a founding principle.
This is exactly what a legal department wants to hear. Not being subject to the rule, but having anticipated it, and being able to document it.
To this are added two substantive guarantees that large accounts demand: data hosting in France, under a strict GDPR framework, and an architecture where one client's data never trains another's AI.
Beyond the AI Act: the two real files
The AI Act is not the only rule of the game, and for a voice agent, they are not even the heaviest ones.
GDPR remains the permanent foundation. An agent that collects a budget, criteria, a situation, and that records exchanges handles personal data. Legal basis, information at the time of the call, retention period of recordings, hosting, sub-processor agreement (Article 28): this is the substantive file, the one a large group scrutinizes well before the AI Act, and the entry ticket to sell to a large account.
Cold calling switches to opt-in on 11 August 2026 (Decree no. 2026-662 of 23 July 2026, applying Law no. 2025-594 of 30 June 2025). Bloctel is over: no more commercial calls to an individual without prior consent that is free, specific, informed and revocable, with proof kept for three years. Important point: this rule targets consumers. B2B stays under the legitimate-interest regime. NoviaMind's outbound prospecting targeting professionals (agencies, networks, executives) is outside the opt-in scope. And inbound, the core of the product, is not cold calling.
In summary
The two major regulatory changes of summer 2026, the AI Act and phone opt-in, do not bite on the core of what NoviaMind does. The transparency required by the AI Act is in place by design. Opt-in cold calling does not concern a B2B use case. What remains is the GDPR foundation, permanent, upheld, and turned into an argument: an AI as serious as you are about confidentiality.
Don't be subject to the rule. Have already complied with it.
This article is for information purposes and does not constitute legal advice. The exact deadlines and terms must be validated with a legal professional before any decision.
Sources
- Article 50 – Transparency obligations (EU Artificial Intelligence Act) — artificialintelligenceact.eu/article/50/
- EU AI Act Transparency Obligations: Preparing for Compliance by 2 August 2026 (Sidley) — datamatters.sidley.com
- EU AI Act: Transparency Obligations Take Effect 2 August 2026 (Cooley) — cooley.com
- Décret n° 2026-662 du 23 juillet 2026 – démarchage téléphonique (Légifrance) — legifrance.gouv.fr
- Démarchage téléphonique : ce qui change le 11 août 2026 — touitou.law