Trustworthy AI For Real Telco Impact

AI in telecom is finally graduating from slide decks to real operational impact, but the jump from pilot to production is where most teams get stuck. I sit down with Guy Lupo from the TM Forum, who leads the trustworthy AI and data mission, to talk about what it actually takes to become an AI native telco and why the industry’s next gains depend less on flashy demos and more on operational proof.

We break down where operators are seeing traction right now, like network fault management, faster mean time to resolve, fewer tickets, and churn reduction, and why those wins correlate directly with clean, structured signals. Then we dig into the uncomfortable middle ground: AI that augments people feels manageable, but AI embedded into tools and workflows raises hard questions about governance, monitoring, and accountability. Guy’s point lands hard: trust cannot be claimed, it must be demonstrated continuously, especially as autonomy increases.

From there, we connect the dots to risk-based regulation and sovereignty. Frameworks like the EU AI Act signal a shift away from checklist compliance toward auditable evidence over time, with telecom increasingly treated as high risk critical infrastructure. We also explore emerging concepts like agent passports, plus why the industry is asking for a shared “agent factory” reference architecture and practical, no regret patterns such as Model as a Service for consistent, governable model access. We close by looking ahead to physical AI and robotics and the surprising telecom advantage: the operational workforce that can install, maintain, and safely support devices at scale.

If you care about AI governance, autonomous networks, agentic AI, and the real-world path to production in telecom, subscribe, share this with a colleague, and leave a review with the one trust gap you want the industry to solve first.

Everyday AI: Your daily guide to grown with Generative AI
Can't keep up with AI? We've got you. Everyday AI helps you keep up and get ahead.

Listen on: Apple Podcasts   Spotify

Transcript

SPEAKER_00 0:00

Hey everybody. Really excited for this conversation today as we talk about how AI and telecom is moving from pilots and promises into real operational impact with a true industry insider and expert from the TM Forum. Guy, how are you? Good morning. I'm well. How are you? Good, great, good to see you. Big fan of the TM Forum and your work. Before we dive into the topics, many topics at hand, uh, how do you describe the TM Forum these days for those who may not be familiar? A little bit about your journey, background, uh, and bio.

SPEAKER_01 0:36

Well, okay. Um, journey, I um I'm um I'll start with the TM Forum and I'll then cover my journey into the TM Forum from the industry. Um, the forum today is doing something miraculous, I would say. It's spearheading the entire AI native telco and how to get into an autonomous enterprise, which is a collection of all the things that you need to do in order to actually get into an AI native. And it's important when we talk about AI native, it is not just bolting on automation or doing the usual. Some things actually need to be refactored, some things need to be rethought and reinvented. The form, if you want to think about it for all the members who listen at the moment, is uh is a largest, I think, telecommunication connectivity alliance with 800 plus members, roughly 135 plus um practitioners, and has three words we need to remember. To get to AI Native and to win that race, you need to be composable, autonomous, and trustworthy. Think of them as three things you need to do. And I head the trustworthy AI and data mission, almost think of us as the last mile of operationalizing all of it. So without trust, it's not gonna go anywhere. Uh, my background is I come from a very long background of uh entrepreneurship in the cybersecurity space, products, and then 15 or 17 years in in the telco in Telstra and NBN Australia, and then I rolled in into the forum as I've been with the forum for volunteering for 10 years as a distinguished fellow. And uh I came on board to make sure everybody can trust their AI.

SPEAKER_00 2:19

Well, that's a bold mission. It's not there today for sure. Uh let's start with some big picture items. Uh, a lot of marketing and hype around AI at the moment, to say the least, with the big IPOs and the investments, and it's uh a whirlwind every day. But where are you seeing real traction in telecom today as we speak?

SPEAKER_01 2:44

The telecom industry is is, as usual, uh a phenomenal innovator and a phenomenal lagger. Uh, I think maybe it's the only industry that always does two things at the two extremes. Um some philosophical discussion for another day, though, but um a lot of adoption and benefits, like 33% uh improvement in mean time to resolve, 15% reduction in tickets, in network fault management, slicing, that's and then customer churn reduction in the area of business. You will see a lot of adoption happens, and there is a direct relationship, mathematical relationship, might I say, between where the data was already clean and structured to the success of the adoption. Faults are very clean. There's a very they're very clean, they're they they come, anomalies churn very clean. It's a signal, both are signals that can be tracked uh very easily. And where the signal was clear, the success was. Where less you can see uh adoption is where the signal is not that clear. Um, and then it the solution is not to get your data right because that that concept of boil the data ocean is a bit gone now because people realize wait a minute, I'm not gonna keep on working to try to get my data right and never get to AI. So it's actually focusing on your minimum varia data, your data product, which is something we've done with the mission, releasing a standard on that. Um, so that's where you can see a lot of a lot of adoption and and and things where you see something that is a bit underclaimed, uh, or actually overclaimed and underdone is the concept of autonomy. There is a lot of use of the word autonomous, but actually, if you follow our benchmark of autonomous network levels and soon to be a trust, and as you get to an autonomous enterprise, if you look into the details, to be autonomous level four is not just bolting on automation with steroids into your usual mess. So that is where we see a little bit of um, shall I say, people claim it, but they don't do it. The blocker the blocker right there is that last mile of the trust. People might be doing pox, might be doing experiments, they want to roll them into production, and then all hell breaks loose, literally. Uh maybe the CEO goes, Am I gonna go to jail? Um, and when people don't or unable to answer, there's no product and production adoption. Simply it stays on the shelf.

SPEAKER_00 5:28

Yeah, really great point. Uh kind of dramatic example there. And speaking of POX, you know, what's mostly experimentation versus operators putting things actually for use in production? Where's the dividing line and um where are we? You know, do you think generally?

SPEAKER_01 5:49

Um I I reckon, I reckon um you you will you will notice um um uh you will notice a lot, as I said, where the data is clear and you get fault managements and you get you get churn and you get bots. Um maybe I'll maybe I'll try to answer that a little bit differently and and see where it goes. The telco and the enterprise in general has become really comfortable in augmenting people. You get a helper, a gen AI. Also, the amount of trust and risk barrier on operationalizing is quite confident because they have you and they invested in you, they did your background check, they know, they know you're not a lunatic. Um, um, so they trust your judgment, so they know you'll be looking after that gen AI. So there is trust there, because you're in the loop still. Then there is taking there's taking AI, which is analytical nature, model nature, not necessarily conversational, and augmenting processes. That's where there's a little bit more unrest. You will see those already in call centers popping in as collecting as much information as possible to give the uh give the uh call center rep, uh give uh give uh your mood, your your weather check of whether you're an angry customer or good customer, have we done good by you or not? That one, the controls they're they're still comfortable with them because it is a is an information-only report. You're not making decisions based on it. Comfort. Are you seeing actually augmentation of technologies with AI, which means somebody has replaced their hardware or have actually bought software that has embedded AI in it and put it in production? Um, nobody knows how to govern this, nobody knows how to monitor this. That's where that's where there's a challenge there.

SPEAKER_00 7:42

So is the challenge around scale or other aspects, or is it one of many? Why is it so hard to scale these pilots? Many, many of which I think have shown some success.

SPEAKER_01 7:54

The pilots are showing success and they give you a great promise. But this is a great question because it actually leads into one of the problems we're looking at at the moment. And then to present this problem and and in a in a way that we all understand it, I I probably don't need to remind you there was a day not long ago in our history, and we used to have this conversation um on a phone bridge wearing pajama and eating popcorn at home, and nobody needs to know what we're wearing, right? Um that's not far away. Uh, and um in a way, in a way, um um came in the revolution of the video, and we all saw an experimentation of video calls in the lab, the CTO department showed to the board, and there was excitement in the birth of video. And then came the horror that followed, and I've I've been in that as I remember it. Is um the CEO goes, Can you put one of these on every desk in the business? Oh my god, you do remember that? Aha, hardware, transit network upgrades, in interoperability, identity, um, devices, supply chain, um calls not working, HLRs. Um, it was an engineering effort to scale. Eventually, we can now have this conversation on video over the internet. But honestly, it was hard, very hard. If I can just convince you to change the word video to agent on every desk, here we go again.

SPEAKER_00 9:33

Great point. I I remember those video calls, you'd have to call it to set up every call. That was just par for the course.

SPEAKER_01 9:41

So you have to really call IT today. You have to call it today to set up inference if you really want to do something heavy. It's absolutely the similar journey. We tend to forget ourselves. But for those of us who have been here a little while, this is just repeating itself. Only this time, the video had trust built into it because it's a point-to-point. This one has no trust built into it. You the more autonomy you give this thing, the less you know whether you're going to jail. And then that's that's where, so there's an engineering scale, and then um understanding that in in December 25 when we started the mission, understanding that the last mile is probably the most important thing at the moment. And then it is we use the tagline of um, we use two taglines, by the way. The first one is we call this operationalization, is the engine of acceleration. To say you must operationalize in an agile way. Operationalize, operationalize, operationalize more, more, more. You will not go to this gym and come out in an hour session, uh, you know, Mr. Mr. Mr. World. It's not gonna happen. It it is a journey of building that muscle. And the second one we tell people that the difference in this day and age, unlike before, trust can no longer be claimed. It must be demonstrated. People would not take your claims. You can claim something you can see in a video. Okay what are you claiming here? You're not gonna do any harm, that your AI is not going to tell the rep to give bonuses to every customer, um, every call, that your AI agent is not gonna go crazy and call the speed test uh 1,000 times per minute and take the network down, cannot claim this. You are in an agent world at the moment.

SPEAKER_00 11:48

Really well said. So we touched on you know infrastructure and IT and data, even culture. What about regulation and governance? What has to change there as AI gets closer to the network?

SPEAKER_01 12:01

We talk about this a lot. Thank you. Um we talk about this a lot. There are there are three drivers that are really driving our industry with this area, and one of them would be regulation. The first one we've just discussed just right now, um, to to bridge from the previous conversation, we call it the AI industrialization. Pretty much um you have to make it in production. It's it's it can't be uh uh a workshop. The second one we call it the agent economy, which everybody sells you agents now. Okay, there's and there's more to come. And uh we can talk about it later as it becomes it because from 100 gram passive device to 15 kilos active device at home. But I'm looking forward to talk to that. And then the last one is the regulation. There's been a worldwide shift in this that maybe it's a tectonic shift and people don't realize it yet. And I I know it all too well. I've been doing my my own in my own personal life. I've been I've been I've been almost finishing my PhD on this, so I've been I've been following this very carefully. The regulators have realized something really important about three and a half years ago. They usually do if you listen to them carefully. Um, they realize that the GDP promise of trillions of dollars, you know how it went from millions to billions of trillions in no time. Okay. By the way, the tokens that we consume are are also going from billions to billions of trillions now. It's I wonder how we make money there. But um, so um they have to protect this future revenue, and they realize also that being regulators and owners of citizens, um, it must be done in a safe way. It it cannot, you cannot harm the people that adopt, knowing that the change is with people, augmenting people, loss of jobs, replacement of jobs. How do they run innovation and not inhibit uh not inhibit innovation and still grow? Um regulation until recent years was all compliance, checklist. Do you have a foul? Yes. Did you did you did you wash your hands before dinner? Yes, okay, I can trust you, you're good. Now you can't be claiming all of this. And um they moved all of them into what's called risk-based regulation. The first one was the uh European Union Artificial Intelligence Act, AIA, followed by the Americans, followed by many countries, and now sovereignty is coming further and even strengthening that. Um, supply chains. What's changing the regulation is a risk-based regulation does not checklist. It's no more like financial regulation that we all know for financial markets. Declare what you're going to do according to this guideline and we'll audit you. And if you demonstrated that you are good with your claim, which means he said, my business will do no harm. I haven't excluded any person, I don't have foul language, and every kid is safe here. There was no incident for three years. We are safe. The auditors will come and check whether you said what you claim. And if you said what you claim and you're continuously saying what you claim, we can trust you. So regulation-wise, all telcos have been deemed high risk and will be audited by the Europeans at least, if not everybody else. Two sovereignty is coming. The sovereign player not only incorporates this but says I am critical infrastructure, and therefore I want to have my own say when you run AI. Some of our members have started implementing with their government. There's one in specifically in Mauritius that is pioneering this at the moment, which is an agent passport. If you want to run an agent in that geography, you got to get a passport, like a birth certificate from the telco and the country, which I really love this idea, and I'm trying to convince the Australians to do the same now. So this is where it's all going. The regulation is becoming intertwined with the telco because there is a fair chance that we will be the ones that actually enforce sovereignty. And it comes together with some sort of uh residency, or do you have a work permit, for lack of a better word, to work in Australia? Be you an agent or a human or tomorrow a robot. Who knows?

SPEAKER_00 16:36

Brilliant. So let's talk about innovation and speed. I mean, when I started in this industry 35 years ago, uh pretty much voicemail and messaging were the two killer apps, the only services that uh I got that were halfway uh interesting as a business professional now. I mean, I'm you know, I'm texting and calling over Starlink uh on T-Mobile in the US. I mean, operators are pushing the speed and innovation boundaries really rapidly. But in AI, it seems are are they gonna have to slow down? And and you know, we can even ask that of Apple, who've kind of kind of missed the boat on AI, at least this initial wave in some regards. So, what do you think? Balancing speed, innovation, control while keeping the momentum.

SPEAKER_01 17:27

If I knew the answer to that question, probably I would be uh Warren Buffett, you know, uh in how to find that right balance. For us in the telco industry, we probably must reinvent, and you'll hear that message coming from us. A lot of our IT. Um we need to actually build that autonomy into the network, and we need to make it in a trustworthy way. That is the big thing. How to do that, how to balance. I think that the role of the forum in this um question is critical more than ever. Standards organizations, um, and I don't call us specifically a standard organization, we're more like a soft standard because we do things by massive agreement and we move forward, we solve problems. Let's call this collaboration and harmonization of things that are safe to do, is critical. So, how to balance that, how to get to that balance. Um, you wouldn't be able to control the capital that flows in, the the promise is too big. Um, you can't stop the experiments, but what you can do is to give experiments a pathway to production with a stable base. And so our mission, specifically for my mission now, because um that's my trustworthy AI and Data mission, um, we have received since uh the last couple of months, and specifically in Mobile World Congress, multiple, multiple requests, every meeting pretty much, have asked us for one thing. Can the TM form do your magic again and give us a reference architecture, telco reference architecture for the agent factory? We have 13, 14, sometimes 16 agent factory diagrams from each and every vendor, consulting company, and um uh and and and uh platform providers. Uh, and we just want to have a safe path to know that if it's infrastructure, data models, agents, operations, what are the agreed technology? Do I have to have a model gateway or do I not? Do I have to have a data semantic layer or do I not? We do this all the time. This is what you do in the form. They've asked us to do it, but they asked us to do it yesterday. The second thing they've asked us to go, once you've done that for us, and that's your blueprint, we call it the AI native blueprint. Okay, um, give us the patterns that are safe to use. What are the real things that are no regret that we can already go and do as a pathway to production? And that's how we'll balance that. Uh, one of those patterns that we've done, uh, and is extremely successful, is called model as a service, MODAZ. Modaz was born in Melbourne, Hill, by Telstra, PWC and Dell, and is now grown to be really big. It's got a catalyst that presents in DTW, Copenhagen, um, and it will demonstrate uh concepts of sovereign inference, intent-based uh model, um, agent asking the platform, um, give me your cheapest model, give me your safest model, without the need to know what model. It would allow our members to onboard, procure contracts, like the good old days, a contract with Gemini, a contract with that host SLMs and service them to agents in a unified way. This is just one example. There are a couple of more patterns that we've done for security, agent-to-agent interaction, some for data products, some for governance and evidence. And we're doing so. My long answer to your short question is finding a way to balance would be to harmonize and standardize not the world, but what we think is no regret, and then move to production and build, build, build on these building blocks. Hopefully that's gonna work. Otherwise, I don't see any other work, uh, any other way.

SPEAKER_00 21:34

Great insight. Uh, so let's shift gears a little bit. We've been all hearing a lot about physical AI and robotics. Oh, fascinating topic. But um, you you've been talking about it in a telecom context. Why and and what does it mean exactly?

(Cont.) Trustworthy AI For Real Telco Impact

SPEAKER_01 21:50

I um there's been something that we are being being where we are, we are now, we always exposed to that, what's happening there at the Edge of iRobot. It's not Skynet yet, but it's never going to be Skynet for those who understand what it is. But we started to think and then engaging in Mobile World Congress, and we had an amazing week in China in May just now with uh with with workshops. We're planning probably to do one in the US as well, maybe one in Europe, uh, about physical AI. And let me let me drop something here that will probably go to the next conversation one day. Um, question number one to be asked from the telecommunication providers that I've been encouraging now, my members. We used to measure our growth, our chief financial officers talk about growth in traffic. Number one, the word bytes are disappearing. China has launched tokens, a mobile plan for tokens, no longer with your bytes. Number two, yes, this is now official. Autonomy is coming. Now, what is autonomy? Autonomy means less tokens. That means you need to less communicate constantly to know what you're doing. You are giving autonomy. Autonomy is autonomy in the physical AI world means that a a piece of embodied intelligence, let's call it the right way, will be operating and cannot be connected all the time. Um, and so this is really interesting. Now, at some point in time, you have to start thinking conversation number one is are we gonna have more in numbers, devices sold or tokens? Which one's gonna grow? Um, I have a personal bet that we'll see more devices sold, as this is the potential, the real potential down the road. That's number one. Number two, you should ask yourself there is a transition, and the transition is not something that we are kind of thinking about at the moment because we're all technology, but the transition is from a hundred gram passive device, your modem, your your TV router, your router at home passive, no moving parts, very cheap, cheap manufacturing, to a 15 kilos or 50 kilos moving complicated machinery that might be at your house or in your factory. 100 gram passive device, 15 kilos dog moving parts. Is this a smart modem that has to be maintained at home? What happens when it breaks? It needs a truck roll. Okay, um, it's got maintenance, all have to be serviced. Um, this this thing are all coming, and we we we are we are this is happening in industry, the industrial well, they have industrial robots at the moment, but at home. The other thing at home, before I tell you what the telco role is, it is something I've heard. Um, and again, these are concepts that are really advanced, but the world um there are things that are called world models. They are not language models, they are models of the language of mother nature. These world models are actually telling you how to interact with nature forces. Why do you do that? They are given to full size or dogs or physical AI. What you and I have learned from since day zero, our neural network in our um room temperature quantum computer, because that's what we are. Okay, you've been you've been wiring yourself when you move your hand in such a way in the train and you might hit somebody, you have body um adjustment. You adjust your force, you adjust your weight. Let me tell you, the robot's gonna smash somebody's face in the train. So they're taking world models and embedding them into the body intelligence of the robot. Body intelligence is a word you'll hear more. Now, why am I telling you all of this? I was asking the question from my telco members, and who is who are the companies on this planet that can put a device at home, connect it, maintain it, track roll when it's broken, have the operational call center, understand home safety, understand health and safety as critical infrastructure and is sovereign to maintain this stuff. This very, very smart, autonomous modem at home. Is that the electricity company? No. Is that the hyperscalers? Probably no. They can ship a box, passive, mate, they cannot operate. Telecommunication companies, and we forget the last word, which is the most important word in our heritage, telecommunication operators. The moat of the telecom industry is its operational workforce that we tend to forget because we like to play with technology, is and might become the actual one that helps roll out physical AI future into the world. My question to the telecommunication industry in this today and in the future are we gonna have a piece of the spy and how?

SPEAKER_00 27:37

Oh, brilliant take. And it's already happening to a tiny degree in healthcare at home and uh remote patient care, remote patient monitoring, robots in the home for elderly aging in home as we're all getting older, including on this show. So it's uh yeah, I'm rooting for success here. And it sounds like science fiction, but very much not. So coming back to uh to Earth, though, for a moment, uh any final message to telecom leaders, maybe CEOs, on why they should attend uh DTW, Ignite, the big uh TM Forum get together in Denmark in just a short number of weeks.

SPEAKER_01 28:16

Um Absolutely. Um message number one is because it's the coolest gathering for the telco because it is a collaborative event which is very different than any other. This is where the real conversation happened. The rubber hits the road. We've always been the rubber hits the road. Pragmatism is our second name. So that's want to have pragmatic conversation on what can be done tomorrow, right now. Come to DTW. The second one is um if you come to DTW and come to the AI summit, uh, which we run from the mission, the autonomous network and composable, we are um going to launch and publish a research we've just finished with our insights and with IBM research, which have helped us do the research about your readiness in terms of trustworthy. And I wouldn't spoil the fun of the results, but I would just say that we are going to point out a very, very serious gap between what people think they are and what is really happening. Um, and this is a call to the industry again. Trust cannot be claimed, no longer claimed. It must be demonstrated and continuously demonstrated, and we are going to lean very heavily into how you can do that last mile so AI becomes reality.

SPEAKER_00 29:37

Brilliant. Well, block mother guests together. And of course, the fact that it's in Denmark in June, one of the most beautiful cities in the world, doesn't hurt. So, everyone head out to Denmark for Ignite in June. And um, great chatting, uh, guy. I really learned a lot. And and thanks everyone for listening, watching. Check out our TV show as well, techimpact.tv on Bloomberg Television and Fox Business. Thanks, everyone. Cheers. Thank you.