- Ericsson and Telstra are extending the autonomous network collaboration they announced in October 2025, moving the focus toward trusted knowledge, semantics and AI decision-making
- Telstra’s Mark Sanders said operators need to move beyond simply adding AI agents to existing automation and give machines enough trusted knowledge to make network decisions safely
- The companies plan to explore how knowledge can be classified, federated across domains and vendors, and ultimately used to support higher levels of network autonomy
ERICSSON OSS/BSS SUMMIT 2026, ZURICH, SWITZERLAND – Ericsson and Telstra are extending their work on autonomous networks, but the next phase isn’t simply about putting more AI agents into network operations.
Instead, the two companies are zeroing in on a thornier problem: What does an AI system need to know before an operator can trust it to make network decisions on its own?
The expanded collaboration builds on work Ericsson and Telstra announced last year around autonomous networks and a proposed “knowledge plane.” This time, the companies are digging deeper into how network knowledge should be structured, classified, shared and trusted across the numerous operational domains and vendor systems inside a telecom network.
For Mark Sanders, chief architect and executive of autonomous networks enablement at Telstra, that distinction is central to moving beyond automation.
“We’re changing the role of our engineers and our people from writing those procedures and scripts to perhaps declaring goals and outcomes and policies,” Sanders said during a press conference at Ericsson’s OSS/BSS Summit in Zurich.
In that model, machines increasingly determine how to achieve the desired outcome rather than humans prescribing every step.
Automation isn’t autonomy
That is a significantly higher bar than using AI to improve an existing automated workflow.
Sanders argued the industry needs to distinguish between increasingly sophisticated automation and genuine autonomy. Simply putting agents around an existing process still leaves much of the underlying decision-making structure intact.
The longer-term goal is for humans to specify an intent or desired network state and allow machines to determine how to get there, within defined policies and guardrails.
That creates a problem: An autonomous system needs information it can trust before an operator can safely delegate those decisions.
“You have humans, people, machines need to coexist off a common understanding,” Sanders said, which is where the companies' knowledge work comes in.
Telstra is exploring a framework that distinguishes between different types of knowledge and the confidence that an autonomous system should place in them. A policy explicitly declared by an operator, for example, isn't the same thing as a conclusion generated probabilistically by an AI model.
Sanders described foundational knowledge as including declared policies and directly observed network conditions, while derived knowledge can include conclusions reached through deterministic calculations as well as learned knowledge generated through AI or machine learning.
The distinction matters because generative AI and large language models are probabilistic. An autonomous network can't simply assume every answer generated by an AI system is safe to execute.
From a paper to the network
The partnership extension also raises an obvious question: What did Ericsson and Telstra actually accomplish during the first year?
Sanders said one of the most visible results was a paper examining the difference between automation and autonomy, including the challenges involved in keeping networks resilient and efficient while safely making changes at scale. However, publication wasn't the only goal.
“This wasn’t just to publish a paper; it was trying to push the industry along with these things,” Sanders said.
The work also fed into industry and standards discussions and a TM Forum Catalyst involving conflict management for intent-based networks, where the companies incorporated an early version of the knowledge-plane concept around policy management and conflicting intents. The expanded partnership's ambition is to make that knowledge model more practical however one big challenge is federation.
Operators such as Telstra have knowledge scattered across fixed and wireless networks, customer systems, products and operational domains, while vendors including Ericsson bring their own product and technology knowledge.
Sanders doesn't expect all of that to be swallowed by one giant centralized system. “We’re not gonna have one all-encompassing big brain,” he said.
Instead, the emerging model is likely to involve localized and distributed stores of knowledge that can be shared or federated when an autonomous system needs to make a higher-level decision.
Taking the human out of the loop
Telstra is already exploring how the approach might work in the network. Sanders described one example in which an operator declares a goal, such as creating a resilient switch fabric with particular capacity and availability requirements. Rather than an engineer designing every configuration and transition step, an AI-based system could determine a network state, evaluate possible plans against policies and operational constraints and construct a transition sequence.
Via the TM Forum catalyst, Telstra is also looking at the approach in the RAN, including energy optimization and load balancing while maintaining customer experience and network stability. Here, the key isn't simply that AI participates in the process. It's that the architecture could eventually shift the human from making individual operational decisions to supervising outcomes.
“We’re trying to move to the human overview, supervising the outcome, and allowing machines to make the decisions,” Sanders said.
As a result, trust, rather than AI itself, is the bigger challenge facing autonomous networking.
The technology is only part of that transition. Sanders said operators also face an organizational challenge as they move toward treating operational knowledge much more formally. “I think knowledge will be seen as a strategic asset,” he said.
AI models, agents and architectures may change, Sanders said, but the operator's accumulated knowledge about its network, policies and business is more enduring.
Read more about Ericsson and autonomous networking on Fierce Network
Verizon, Ericsson dial up 6G in 7 GHz field test