AI governance: Can regulation keep pace with autonomous networks?

Details: James Barton; Category: Wireless Networks; 10 June 2026; 635 views

Regulators, famously, move slowly – while AI models are evolving at an unfathomable speed, outpacing the speed of governance whether intentionally or not. The potential benefits of AI-driven autonomous networks are vast, but so is the possibility for disaster if AI agents are able to act without accountability.

The EU AI Act has emerged as something of an initial standard-bearer in terms of AI governance, and while the idea of regulating AI regionally may seem futile, there is global recognition that an international standard should be reached – even amongst markets eager to adopt AI as quickly as possible. In this week’s feature, we explore the state of play in AI governance and how this impacts emerging markets.

The EU AI Act and the search for standards

Nik Kairinos, CEO and Co-founder of AI monitoring platform RAIDS AI, notes that while the EU AI Act is a complex piece of legislation, many of its original aspirations had to be pared back given the speed at which AI advances. The AI of today is unrecognisable to the AI of two years ago – and this will apply equally two years from now. Given that regulators move at very slow speeds, this represents enormous challenges in determining what requires regulation.

Kairinos explains that sources who worked on the EU AI Act at one point were attempting to argue that AI could not be considered legal if it could not explain itself – a difficult proposition given the black-box nature of deep learning.

“Not even the best AI researchers in the world truly understand how the model works - they understand the architecture, the algorithms, the different functions, and so on, but not truly how it maps things within the neural nets”, explains Kairinos.

Describing the EU AI Act as an initial drop in the ocean, Kairinos notes that it has triggered numerous regulations globally, each focused on different aspects of AI. He observes that South Korea is perhaps taking the most similar approach to the EU Act, while China is taking an agile approach that is largely focused on content governance.

This growing regulatory patchwork has also accelerated interest in common standards. Kairinos points to ISO 42001, which he describes as the ‘Rosetta Stone’ of AI regulation.

“At enterprises, governments, and others that are acquiring AI at a very quickly growing rate, they want ISO42001 certification, because to try and keep up with all the different regulations is impossible.”

Why monitoring matters

Kairinos notes that one common theme across approaches to AI governance is a focus on real-time monitoring. AI models are no longer static systems; they evolve through exposure to new data and increasingly through self-improving capabilities. As a result, point-in-time assessments quickly become outdated.

“One of the central themes is this notion of real-time monitoring”, says Kairinos, “because the only thing that can keep up with AI is AI. There’s got to be an audit trail when things go wrong. Explainability has to be a component of it. Originally they were trying to go for this whole idea that the model has to explain every little decision it makes, which is just not possible.”

For telecom operators progressing towards more autonomous networks, monitoring and measurement will remain essential as increasingly complex agentic AI systems are introduced into the network.

TM Forum’s Executive Vice President Andy Tiller explains how members working on TM Forum’s autonomous networks project have developed tools that can measure the level of autonomy across a wide range of scenarios.

“They define the level of granularity”, notes Tiller. “We’re measuring autonomy - not for the network, that’s just far too broad - you have to build it down to a level of fault management on the RAN or energy efficiency optimization on the transport network.”

The road to Level Five autonomy

In terms of autonomy, the current industry target is Level Four, where AI takes over decision-making in specific operational scenarios. Beyond that lies Level Five, where multiple autonomous systems operate together across the network.

Tiller highlights the importance of coordination between agentic systems. An AI application optimising energy efficiency, for example, could potentially conflict with another system designed to maximise customer experience.

“That’s where we get into this domain, where you have to coordinate across different agentic systems that might have different intents built in”, explains Tiller. “That’s where the work is at the moment - we’ve got some quite mature tools already around specific, high value scenarios, and we’re exploring now how those scenarios can be joined up and integrated into a more coherent thing, which takes you beyond Level Four towards Level Five, which is where everything’s joined up and run by machines.”

Asia’s faster path to AI adoption

Interestingly, Tiller notes that operators in Asia are largely ahead of their western counterparts, noting that China in particular is well advanced, having anticipated the shift towards autonomous networks and accordingly adopted AI rapidly.

“China tends to roll out new technology first in a province that would typically be the size of a European country. Zhejiang or Guangdong on the East Coast, they’ve got some fantastically advanced automation solutions for particular scenarios in those provinces”, notes Tiller.

He underlines that these solutions have been exported globally to operator customers in Latin America, the Middle East and other parts of Asia where Chinese vendors such as Huawei and ZTE have strong supplier relationships. Tiller adds that Ericsson and Nokia also have highly advanced deployments across customer bases in Asia and the Middle East.

Africa, Latin America and the Middle East are characterised by a greater reliance on suppliers than many large western operators, which often possess significant in-house software engineering and AI capabilities. Emerging market operators frequently depend more heavily on vendors such as Huawei, Ericsson, Google or Microsoft to deliver AI solutions.

This supplier-led model can accelerate deployment and adoption, but it also presents challenges when operators seek to integrate multiple applications and vendors into a unified autonomous network environment.

“The moment you want all this stuff to join up, you’ve got to start layering it onto a common foundation and that’s where you either accept you’re going to be locked in, or you say ‘we’ve got to build on some standard foundations.”

East versus West

So what drives this difference between East and West? For Tiller, many of the factors are cultural. While regulation plays a role, the rate at which new technologies are adopted across China is often faster than elsewhere, whether in autonomous vehicles, mobile internet or digital payments.

He notes that China’s centralised investment model has enabled major technology initiatives, but argues that data readiness has also been critical. AI deployment depends heavily on breaking down data silos, preparing datasets and making them usable for training, fine-tuning and inference.

According to Tiller, Chinese operators were already well advanced in these areas before the current AI boom. Combined with the global reach of vendors such as Huawei and ZTE, this has helped accelerate deployment across multiple emerging markets.

“Whilst I would say Asia is ahead, it’s not just because the Asian operators are more advanced - it’s also because they’ve adopted the technology, whether it’s east or west. It’s coming more quickly, and I think what characterises Asia is this fast moving; the Western operators tend to be a bit more careful about introduction and new technology, perhaps a bit slower and more deliberate about it.”

Trust remains the key challenge

The EU AI Act – and similar regulations – are often viewed by technology companies as barriers to faster adoption. Tiller urges caution. Achieving Level Five autonomy requires operators to trust machines with increasingly critical operational decisions.

To reach that stage, guardrails, controls and security mechanisms must be built into the foundations of autonomous networks. Agentic systems cannot be given unrestricted access to models and datasets; governance must be applied consistently across the environment rather than separately for every application.

“When you start to put all that together, you’ve got to build in a very high level of trust”, says Tiller. “I wouldn’t say that regulation is the thing that’s going to make or break it. That’s going to be one bar that could be high or less high, but there’s a lot of other bars as well that every operator is going to want to comply with in order to trust the AI to take over operations.”