An example of secure and fluid mobile use of generative AI with compliance analysis by video at the MWC

• 5G
• AI
• Connectivity
• Data
• Work

Share on Twitter : An example of secure and fluid mobile use of generative AI with compliance analysis by video at the MWC (New window) Share on Linkedin : An example of secure and fluid mobile use of generative AI with compliance analysis by video at the MWC (New window)

The year is 2025. A network field engineer wants to field test that a router is properly connected. They use their mobile device to transmit a video stream of the equipment. On their spoken instruction, software trained to apply a written specification, along with a video analysis, decides whether or not what is filmed is compliant. Its secret formula: a combination of AI and features specific to 5G networks — slicing and installation of 5G relays on company premises (edge computing).

This software is called Live Intelligence. A few months after its launch in November 2024, the prototype of a more advanced version is being exclusively unveiled, featuring real-time video inference and mobility, owing to the ever-changing uses and technology of AI. The unveiling will take place at MWC (the Mobile World Congress) 2025, where Orange is running a demo of the secure uses of generative AI on the go, made possible by 5G standalone and generative AI. To its knowledge, the carrier will be the first in the world to demonstrate live the added value of a conversation with generative AI (with real-time video inference) thanks to 5G slicing. Orange is able to offer these new features thanks to its dual expertise regarding networks and AI.

The carrier is demonstrating the added value of a conversation with generative AI (with real-time video inference) thanks to 5G slicing.

From Dinootoo to Live Intelligence

In 2023, Orange launched an internal service called Dinootoo, so that all its employees could discover the potential of LLMs and make good use of them, in a way that suits their professional needs and without any risk to the company in terms of data privacy. The solution is a uniting force, according to Antoine Maisonneuve, Product Team Leader Live Intelligence at Orange: “Dinootoo is one of the largest generative AI platforms operating in Europe today. Over 60,000 unique users have used the solution since its launch. Its use cases are constantly becoming more diversified, sophisticated and adjusted to the specificities of each business. In particular, these include the analysis of confidentiality agreements by our teams of lawyers or calls for tenders by sellers. And we are continuing to enhance the platform from a functional point of view in order to meet these high expectations, as can be seen from the upcoming integration of Mistral AI’s Codestral code-generation model”.

With Live Intelligence, Orange has leveraged this expertise in scaling up such an application, combined with an optimised methodology, offering a similar multi-LLM solution designed for the B2B market, with features that are improving from month to month. Two versions are available to date:

• Live Intelligence Trust — a sovereign generative AI solution, hosted on the Cloud Avenue platform, operated with GPUs and by Orange Business experts. It is primarily aimed at communities and companies seeking a maximum level of confidentiality.
• Live Intelligence Open — a more general version launched in beta in March 2025. It uses a partner cloud and aims to protect against the phenomenon of shadow AI, i.e. the uncontrolled use of company data through generative AI by employees.

In both cases, security is at the heart of the value proposition. From a technical and operational perspective, Trust ensures end-to-end security of customer data and, from a contractual perspective, Open guarantees that data is not reused for retraining LLM models, in particular. The challenge of security goes hand in hand with the accessibility of the solution. Aurelle Arber, Operational Marketing Manager at Orange Business, explained: “The main aim is to increase the use of generative AI in a secure way by positioning ourselves as a facilitator and a trusted player. The Live Intelligence range offers tools to activate and simplify usage, such as wizards and preconfigured action libraries. But beyond the offering, we also assist customers through appropriate training, change management plans etc. This support is essential to prevent the company from exposing itself to risks, but also to allow it to show the benefits of AI to their full extent”.

Networks in the era of AI

Fluid, conversational interaction with such a tool, which is being demonstrated at the MWC, gives rise to many uses that will tap into carriers’ know-how. This on-the-go implementation is a real step forwards, involving new challenges and new responses around security and network operations. “AI traffic is different from standard traffic and one of the specific features is that it is uplink, while networks are traditionally calibrated for downlink traffic”, Taoufik En-Najjary, Data Scientist at Orange, said. “So we need a specific connectivity model, which we build by combining three technical components. To start, it’s a question of configuring two distinct and divided spaces on the user’s mobile device — one for personal use and another secured for professional use. The second step is to dedicate a 5G slice to this professional space, in order to allocate and prioritise bandwidth capacity on the AI application. Finally, we draw on the local breakout process introduced with 5G to ensure that sensitive data is not transmitted over the Internet by moving a fragment of 5G core to the customer’s premises for end-to-end control and fast data flow”. As a result, employees can exchange data with AI confidentially and benefit from a high level of responsiveness, essential to making emerging audio and video interactions work with AI.

The MWC demo shows that telcos have strong assets for this new type of traffic, which is arriving in networks. In the future, Orange plans to keep on innovating to make sure that it has responsible and inclusive networks for AI.