AI Voice Agents in Italy: Garante & GDPR Compliance 2026

AI voice agents deployed in Italy must comply with four overlapping regimes: GDPR (Regulation (EU) 2016/679), the Italian Codice in materia di protezione dei dati personali (D.Lgs. 196/2003 as amended by D.Lgs. 101/2018), the EU AI Act, and Italian telemarketing rules anchored on the Registro Pubblico delle Opposizioni (RPO) — all supervised by the Garante per la protezione dei dati personali. The Garante is the most active EU DPA on AI matters, having issued the first-ever EU-wide temporary ban on a generative AI service in March 2023.

Italy is one of the toughest EU jurisdictions for outbound telemarketing. The Garante combines GDPR enforcement with sector-specific decisions on telemarketing fraud, automated systems, and AI training data. Any AI voice deployment touching Italian residents must layer all four regimes from day one.

What Is the Italian Compliance Framework for AI Voice?

Italy stacks the GDPR baseline, the Codice Privacy national overlay, sector telecom rules anchored on AGCOM and the RPO, and the EU AI Act transparency layer. Each adds requirements the others do not. The Garante coordinates enforcement on all four for data-protection breaches; AGCOM handles pure telecom violations; the AGCM (Autorità Garante della Concorrenza e del Mercato) can add consumer-protection sanctions if calls are deceptive.

The Italian framework is more stringent than the EU baseline in three places: (1) consent for marketing must be explicit and freely given, with separate consent for different processing purposes; (2) outbound calling has a 15-day RPO rescreen requirement; (3) employee monitoring under Article 4 of the Workers' Statute (Statuto dei Lavoratori, L. 300/1970) requires union agreement before AI-assisted call QA can be deployed (Source: Gazzetta Ufficiale L. 300/1970).

Who Is the Garante and What Does It Regulate?

The Garante per la protezione dei dati personali is the Italian supervisory authority under GDPR Article 51. Established in 1997 by the predecessor of the Codice Privacy (L. 675/1996), the Garante predates GDPR by two decades and has built a deep doctrinal base on commercial communications, automated decision-making, and biometrics.

The Garante is uniquely activist on AI. In March 2023 it issued the Provvedimento 9870832 temporarily blocking ChatGPT for Italian users — the first EU-wide regulator action of its kind. The subsequent 2024 EUR 15 million sanction (Provvedimento 10085455) confirmed the Garante's position that generative AI services must satisfy lawful basis, transparency, and minor-protection rules even for training data.

For AI voice agents specifically, the Garante's 2018 Provvedimento generale call center remains binding. It requires Italian transparency disclosures at call start, recording-purpose specification, retention limits, and worker notification.

How Does the Italian Codice Privacy Overlay GDPR?

The Codice (D.Lgs. 196/2003), as amended by D.Lgs. 101/2018 to harmonise with GDPR, adds Italian-specific rules:

• Article 130 — Direct marketing communications. Automated calling systems (including AI voice) for marketing require prior explicit consent (opt-in). Calls placed manually are subject to the RPO opt-out regime.
• Article 132 — Traffic data retention. Italian telcos must retain traffic data for 24 months for law enforcement purposes. Voice agent platforms that interact with the telco layer must understand the data-flow implications.
• Article 2-quaterdecies — Children's consent. Children must be 14 or older to give valid consent for information society services.
• Article 2-septies — Special categories. The Garante can adopt additional measures for processing of sensitive data, including health data heard during calls.
• Article 166 — Sanctions. Italian-specific administrative sanctions apply on top of GDPR Article 83 fines, including the power to order the blocking of unlawful processing.

What Lawful Basis Applies to AI Voice Calls in Italy?

The Garante's position on lawful basis for voice agents in Italy:

The Italian distinction between "automated" and "manual" outbound marketing is critical and easy to misread. Under Codice Privacy Article 130(1), automated calling systems (which include AI voice agents placing outbound calls) require opt-in consent — the RPO opt-out is not sufficient. The Garante has been explicit that pre-recorded messages and AI-driven calls fall into the automated category. Human-dialled, human-spoken calls are the manual category and can rely on RPO opt-out under Article 130(3-bis).

In practice, the safest pattern for AI outbound marketing in Italy is double-screening: (a) consent on file from the recipient AND (b) RPO screen at most 15 days before dialling.

How Does the Registro Pubblico delle Opposizioni Work?

The Registro Pubblico delle Opposizioni (RPO) is Italy's public opt-out register for telemarketing and direct mail, operated under DPR 26/2022. Originally limited to landlines, it was extended in 2022 to include mobile numbers and all published numeric directories.

Telemarketing operators (including AI voice deployments) must:

• Register on the RPO portal as a soggetto utilizzatore (data user).
• Submit call lists for screening before any outbound campaign.
• Rescreen at least every 15 days for any continuing campaign — this is shorter than Spain's 30-day cadence.
• Honour consumer revocations made directly to the operator within 24 hours.
• Retain screening records for at least 5 years for Garante and AGCOM audits.

Failure to screen against RPO is sanctioned by AGCOM up to EUR 700,000 per infraction, in addition to potential Garante GDPR sanctions. Coordinated AGCOM-Garante actions in 2023-2024 against telco resellers produced cumulative seven-figure penalties.

What Are Italian Call Recording and Consent Rules?

Disclose Recording at Call Start

The Garante 2018 call-centre provvedimento requires that any call recording be preceded by an Italian-language disclosure naming the controller, the specific purposes of recording, the lawful basis, and where the full privacy notice can be consulted. Generic "questa chiamata può essere registrata" is not sufficient if the purpose is not specified.

Layered Information Notices

Like Spain, Italy accepts and encourages layered notices: a short first layer delivered on the call (controller name, purpose, full-notice URL, data subject rights channel) plus a complete second layer hosted at the URL.

Worker Monitoring (Article 4 Statuto dei Lavoratori)

If call recordings are used for staff performance monitoring or AI-assisted QA scoring, Article 4 of the Statuto dei Lavoratori (L. 300/1970) requires either (a) prior collective agreement with the trade union representatives, or (b) authorisation from the Ispettorato Nazionale del Lavoro. AI voice deployments that monitor agent performance must navigate this before going live.

Retention

The Garante's standard expectation is 30 days for QA recordings, extended only when retention is justified by a specific business need (dispute resolution, regulatory obligation). Permanent retention is disallowed under Article 5(1)(e) GDPR data minimisation.

How Does the EU AI Act Apply to Voice Agents in Italy?

The EU AI Act (Regulation 2024/1689) entered into force on 1 August 2024 and applies directly in Italy without national transposition. Italy is preparing complementary national legislation under the DDL Intelligenza Artificiale approved by the Council of Ministers in April 2024.

For Italian AI voice deployments:

• Article 5 — Prohibited practices. Subliminal manipulation and exploitation of vulnerabilities are banned.
• Article 50 — Transparency. Italian-language AI disclosure at call start is required. The Garante has consistently held that this disclosure must be unambiguous and made before any data exchange.
• Article 6 + Annex III — High-risk. Voice agents used for creditworthiness, employment, or essential-services access fall into the high-risk category, triggering full conformity assessment.

Italy has designated the Agenzia per la Cybersicurezza Nazionale (ACN) and AgID (Agenzia per l'Italia Digitale) as competent authorities for non-personal-data AI oversight. The Garante retains authority where personal data is involved — which is essentially every voice agent deployment.

What Has the Garante Decided About Automated Calling?

Notable Garante decisions relevant to AI voice agents:

• Provvedimento 9870832 (2023) — Temporary block on ChatGPT for lacking lawful basis for training-data processing and inadequate age verification.
• Provvedimento 10085455 (2024) — EUR 15 million sanction on OpenAI after follow-up investigation.
• Provvedimento 9039945 (2018) — EUR 2.0 million combined sanction against four telecoms operators for systematic violation of marketing-call consent rules.
• Provvedimento generale call center (2018) — Standing guidance on call-centre operations: transparency at call start, recording purpose, retention, worker notice.

Across these decisions the Garante emphasises three controls: (a) lawful basis must be valid at the moment of the call, not retrofitted; (b) transparency notices must be delivered in Italian, in plain language, before any processing begins; (c) automated decision-making (Article 22 GDPR) requires meaningful human review, not rubber-stamped logs.

Garante-Aligned Vendor Checklist

Before contracting an AI voice vendor for Italian deployment, verify:

• EU/EEA data residency for raw audio, transcripts, and any model inputs.
• GDPR Article 28 Data Processing Agreement available in Italian on request.
• Native Italian language support (TTS + STT + dialogue model) — not a machine-translated overlay.
• Configurable layered transparency notice at call start (controller name, purpose, full-notice URL).
• Italian AI-disclosure phrase configurable and consistently spoken on every interaction.
• Native RPO list screening for outbound campaigns with 15-day rescreen cadence.
• Configurable retention with default aligned to Garante 30-day expectation for QA recordings.
• Data subject access and erasure workflows that complete within the GDPR 30-day window in Italian.
• Documented incident response with sub-72-hour Garante notification capability.
• Article 22 GDPR safeguards — meaningful human review for any consequential automated decision made on a call (loan eligibility, fraud flag, service termination).
• Documented support for Worker Statute Article 4 — staff notification of QA monitoring before go-live.

Frequently Asked Questions