LONDON – Network leaders in the telecom domain argued that cybersecurity for AI needs to be as much of a focus as progress in autonomous networks.

Gabriela Styf Sjöman, MD for research and networks strategy at BT Group, told the audience at this week's FutureNet World: “We don't often speak enough about AI for cyber,” noting the cybersecurity space is shifting away from a defense problem to an adversarial AI competition.

“We need to use AI to fight AI,” Styf Sjöman said. “Today, 80% of phishing attacks are AI-generated. So [BT] is focusing a lot on using AI for defense and response [as well as being] able to use AI for recovery.”

Joe Hogan, networks CTO at Amdocs, added the firm is seeing “some really interesting cyber use cases” around server farms that “aren't servers or phones.”

“We've got a really interesting use case that tackles that; we're using bandwidth management, and seeing opportunities to monetize the network using AI for bandwidth management that we wouldn't have been able to do in the past,” Hogan said. “This means utilizing our policy systems and our charging system to ensure we're paying for that.”

VodafoneThree chief network officer Andrea Donà added that security would benefit from an overall domain-centric approach to the network. Instead of a single chief data officer dictating everything, he argued, data use and AI initiatives should be owned within each domain, spanning network, security, packet core, and beyond.

Donà explained such an approach would help keep the context of that domain; otherwise, the data becomes “useless." Such an approach would, however, require operators to simplify legacy operations support systems (OSS) and improve observability and data readiness.

“You also have to have a platform that does your data governance, your large language model (LLM) life cycle management, your machine learning, in real time if you like, and which then separates consumer data from network data from OSS data, so that it can be correlated, managed in a secure way,” Donà said. “Otherwise, you're going to get yourself very quickly, a bit of a mess if you're not sure what you're getting from what source.”

According to the VodafoneThree chief, this can help with real-time testing of AI-driven defense mechanisms.

“Being proactive rather than being reactive to it can only come to life if you ground that accountability in the domain-specific area. Because a network engineer isn’t a security engineer, they won't be able to do that," Donà said.

In a follow-up chat with SDxCentral, Sam Keys-Toyer, global head of portfolio for network managed services at Ericsson, gave the vendor view on security in autonomous networks, arguing it needs to be baked into AI within the system.

"You've effectively got entities which have permissions and skills, and they will do certain work and things like that," Keys-Toyer said. "So rather than make security kind of too top-heavy, we need to effectively take what we know and apply it by design into what we're doing so that we have that natural flow through in terms of the integrity of the working environment, which is going to have more 'digital employees,' as well as your human, physical employees."

"From a productivity perspective, security can leverage some of the capabilities that we can provide: how fast we can react to threats, how fast we can look out for threats into the future," Keys-Toyer added. "We can't be complacent."