Deutsche Telekom unveiled a voice AI assistant built directly into its telephone network, requiring no application download, no compatible device, and no manual configuration from the caller. The product - called the Magenta AI Call Assistant - was demonstrated live at Mobile World Congress in Barcelona on March 2, 2026, marking what the company describes as a world premiere in network-embedded AI for consumer phone calls.

The announcement places Deutsche Telekom at the centre of a rapidly accelerating race among telecommunications operators and technology companies to embed artificial intelligence into communication infrastructure rather than confining it to standalone applications. The company presented the assistant as part of its broader AI strategy at MWC Barcelona, where it occupies booth 3M31 in hall 3 from March 2 to 5, 2026.

A call, not an app

The mechanism is deliberately simple. During any phone call, a Deutsche Telekom customer says "Hey Magenta" to activate the assistant. It then joins the conversation in real time, without requiring either party to install software beforehand or own particular hardware. According to Deutsche Telekom, this design removes what the company characterises as the main barriers to AI adoption: complexity, exclusivity, and device dependency.

"Our customers will get AI services like translation right in their calls," according to Abdu Mudesir, Board Member for Product and Technology at Deutsche Telekom. "With our Magenta AI Call Assistant, we are the first in the world to offer these network-based AI functions. We remove barriers - no apps, no special devices, no technical complexity. AI becomes simple, intuitive and available to everyone."

The initial feature set covers three areas. First, live translation operates across languages, allowing both parties in a call to speak and hear in their respective languages simultaneously. Second, conversation summaries help callers retrieve important follow-ups after a call ends. Third, contextual assistance provides answers to questions or pulls relevant information during the conversation itself.

These capabilities are scheduled to reach Deutsche Telekom customers in Germany starting later in 2026. Over the following 12 months, the company plans to extend support to up to 50 languages. Further services - beyond translation, summarisation, and contextual assistance - are described as forthcoming.

ElevenLabs and Radisys: the technical architecture

The Magenta AI Call Assistant was developed in partnership with ElevenLabs, the AI research and product company known primarily for its voice synthesis and agent models. ElevenLabs provides the voice AI models and agent infrastructure that power the real-time capabilities inside live calls. The technical integration at the network level was carried out by Radisys, a company specialising in open telecom solutions.

This three-party structure - Deutsche Telekom providing the network and customer base, ElevenLabs supplying the voice AI layer, and Radisys handling the technical embedding - illustrates a model of partnership rather than vertical integration. Neither Deutsche Telekom nor ElevenLabs has built the full stack alone.

Mati Staniszewski, co-founder and chief executive of ElevenLabs, described the scope of the ambition in a LinkedIn post published on March 2, 2026. According to Staniszewski: "Together with Telekom, we are redefining what a phone call can be. For the first time, AI agents will be embedded directly inside a call - translating languages, integrating context, and taking action in real time. With our AI agents and voice, and Telekom's network expertise, conversations will no longer be limited by language barriers or admin."

Staniszewski also noted in the same post that the collaboration between the two companies began approximately two months earlier, when ElevenLabs and Deutsche Telekom announced work to bring voice agents into Deutsche Telekom's customer service operations. Today's announcement represents an expansion of that earlier arrangement into peer-to-peer consumer calls - a meaningfully different technical and commercial context.

What the assistant can do during a call

Deutsche Telekom's announcement outlines several practical scenarios in which the assistant is designed to operate. When two callers are deciding where to meet, the assistant can suggest nearby restaurants matching both parties' schedules and preferences. When planning a trip, it can compare travel options and make recommendations tailored to whether the travellers are families with children, couples, groups of friends, or individuals. When coordinating appointments, it can check availability and structure next steps.

The real-time translation function is perhaps the most technically demanding of the initial features. It allows each person in a call to speak their own language while the other hears a translated version, without either party pausing to use a separate application or service. The practical implications for cross-border business calls, for migration contexts, or for customer service interactions across language barriers are substantial.

Looking further ahead, Deutsche Telekom describes a future in which the assistant moves beyond information retrieval into action. According to the announcement, if it becomes clear during a conversation that a reservation is needed - for a restaurant, a doctor's appointment, or travel - the AI can make the booking directly, fill out forms, ask clarifying questions such as "What time suits you?" or "Window or aisle?", and then document the confirmation. The assistant is envisioned as transforming conversations into completed actions.

Trust and consent are explicitly addressed in Deutsche Telekom's announcement. According to the company, the Call Assistant processes spoken content only after the customer explicitly activates it during a call. Without that activation, no conversation content is stored or analysed. Customers must opt in before use. Crucially, all call participants are notified when the assistant becomes active - a design choice the company ties to European data protection standards.

This approach to consent - explicit activation, participant notification, no background processing - is significant given the regulatory environment in which Deutsche Telekom operates. European data protection authorities have been publishing increasingly detailed guidance on the compliance requirements for AI systems that process personal data. Spain's data protection authority published a 71-page guide in February 2026 examining how agentic AI systems create privacy risks that existing frameworks were not designed to handle, covering prompt injection, memory risks, and automated decisions. The Dutch data protection authority issued a formal warning in February 2026 against open-source AI agents, specifically citing the risks of granting AI systems full access to communications and files.

Deutsche Telekom's opt-in design and participant notification requirement appear calibrated to pre-empt these concerns. The company frames the architecture as reflecting "European data-protection standards and Deutsche Telekom's commitment to responsible AI." Whether regulators will scrutinise the specifics of how spoken content is processed, stored temporarily, and used to train or improve models remains to be seen.

The broader shift: AI embedded in infrastructure

The significance of today's announcement is not only what the Magenta AI Call Assistant does, but where it sits architecturally. By embedding AI at the network level rather than at the device or application layer, Deutsche Telekom is making a claim about where AI's future home in communications will be.

This is a distinct approach from the dominant model, in which AI capabilities arrive through applications - a dedicated app, a browser extension, a smartphone assistant activated through a device's operating system. Staniszewski's LinkedIn post made the contrast explicit: "Access to intelligence usually means downloading a new app, visiting a new website, or relearning a new interface. Here, we're doing it in the most natural way - through voice - and in the flow of how you already communicate with another person - a phone call."

The advertising and marketing industry has been tracking the expansion of AI agents into operational infrastructure throughout the past several months. The IAB Tech Lab announced in January 2026 an agentic roadmap extending OpenRTB and existing standards with modern execution protocols, while Yahoo DSP integrated agentic AI capabilities directly into its demand-side platform on January 6, 2026. Deutsche Telekom's announcement follows this pattern of embedding autonomous AI into existing operational layers rather than building separate AI-native surfaces.

Mudesir articulated the company's strategic direction plainly in the announcement. "AI makes telephony even more valuable. And with Telekom, phone calls become more than just voice or video. We are breaking new ground with strong partners like ElevenLabs. Our new AI service will be embedded into our network. It overcomes language barriers, helps capture information, and coordinates appointments. This is how we make calls more trustful and effortless." His framing for the longer term: "This is another step on the way to becoming an AI-first telco that sets the standard and enables secure AI experiences."

What this means for marketing and advertising

For marketing professionals, the Magenta AI Call Assistant raises immediate questions about how AI-mediated phone calls will affect consumer behaviour, measurement, and targeting. Phone calls remain a significant conversion channel for categories including local services, financial products, healthcare, automotive, and travel. The introduction of AI summarisation and action capabilities into calls could materially change how those conversions are logged, attributed, and followed up.

The real-time translation function has particular relevance for businesses operating across European language markets. A company running campaigns in Germany, France, and the Netherlands currently faces operational friction when a consumer calls from across a language boundary. An AI translation layer embedded at the network level could reduce that friction substantially - though it also introduces questions about which entity holds the transcribed or summarised content of those calls, and under what data retention policies.

The consent and notification model Deutsche Telekom has described could also become a reference point for how call-based AI services are regulated. European regulators have been actively shaping the compliance environment for AI agents, with proposals under the EU's Digital Omnibus initiative seeking to reshape personal data definitions and establish new legal pathways for AI processing. How a network-embedded call assistant maps onto those evolving frameworks - and specifically whether call summaries or action logs constitute personal data requiring protection - is a question regulators will likely address as the service scales.

The partnership model itself also merits attention. Deutsche Telekom is not building voice AI models independently; it is licensing ElevenLabs' technology and integrating it via Radisys into its network layer. This means the AI capabilities available to German consumers later in 2026 will reflect ElevenLabs' model development roadmap as much as Deutsche Telekom's own decisions. For advertisers considering how this channel might eventually be used for customer interaction, understanding the AI provider behind the network layer will matter.

The MWC Barcelona demonstration today follows a pattern of live event reveals designed to signal strategic intent. Whether the initial German rollout reaches the planned scale - 50 languages within 12 months of launch - will depend on technical delivery, regulatory clearance, and consumer adoption. The announcement sets a specific, measurable target against which Deutsche Telekom's execution can be assessed.

Timeline

Summary

Who: Deutsche Telekom, together with ElevenLabs (AI voice model provider) and Radisys (network-level technical integrator), announced the Magenta AI Call Assistant. Abdu Mudesir, Board Member for Product and Technology at Deutsche Telekom, and Mati Staniszewski, co-founder and CEO of ElevenLabs, were the key executives cited.

What: The Magenta AI Call Assistant is a voice AI service embedded directly into Deutsche Telekom's telephone network. It activates during a live phone call when a user says "Hey Magenta," requiring no app or special device. Initial features include real-time translation across languages, conversation summaries, and contextual assistance. The system operates on an explicit opt-in basis, with all call participants notified upon activation.

When: The product was unveiled and demonstrated live at Mobile World Congress in Barcelona on March 2, 2026. Availability for Deutsche Telekom customers in Germany is planned for later in 2026, with support for up to 50 languages expected within 12 months of launch.

Where: The announcement was made at MWC Barcelona, hall 3, booth 3M31. The service will initially be available in Germany, with international expansion anticipated. Technical integration at the network level was conducted by Radisys.

Why: Deutsche Telekom frames the assistant as part of its strategy to become an "AI-first telco," embedding AI into existing communication infrastructure rather than requiring consumers to adopt separate applications. The goal, according to the company, is to make AI services accessible to all phone users regardless of device or technical literacy, while overcoming language barriers, reducing administrative friction in calls, and transforming spoken conversations into completed actions.

Share this article
The link has been copied!