Autonomous Networks Level 5 — Trusted Decisions, Self-Evolving Networks

Imagine a network that runs itself end to end, and that you can trust to do so. Level-5 autonomy is not more automation. It is a network whose AI takes high-quality decisions without a human in the loop, knows the limits of its own judgement, can explain and prove why it acted, and keeps evolving while staying within a safe, verified envelope.

We’re looking for bold technical innovations that make this level of autonomy achievable: world models and network digital twins that let the network test an action before taking it, methods that make critical behavior structurally impossible, explainability and provenance that turn trust into an engineering property, and ways to judge the quality of what a self-evolving network creates on its own.

What We’re Looking For

Here are some priority areas, though we welcome bold, creative ideas beyond this list:

• High-Quality Decisions: How could an autonomous network take consistently high-quality decisions at scale, using world models, network digital twins or learned simulators to evaluate an action before executing it in the live network?
• Safe by Design: How could critical or irreversible behavior be prevented structurally rather than detected afterwards, for example through safety envelopes, verified action spaces, shielding of learning agents, blast-radius containment and guaranteed rollback?
• Knowing When Not to Decide: How could the network quantify its own uncertainty, recognize situations it has never seen, and escalate to a human as a calibrated decision rather than a fallback?
• Explainable and Provable Decisions: How could every autonomous decision be traced to its cause, explained in operational terms and audited afterwards, so that trust becomes measurable rather than assumed? What role could telco foundation models play in making decisions transparent?
• Robust Against Manipulation: How could autonomous decisions stay sound when the telemetry, models or agents they rely on are corrupted or attacked, for example through poisoned data, adversarial inputs or compromised agents holding write access to the network?
• Self-Evolving Networks: How could a network go beyond executing decisions and evolve itself, deriving new configurations, capabilities or service definitions on its own, while remaining reversible and inside a verified envelope?
• Judging Auto-Innovation: When a self-evolving network autonomously produces an artefact, such as a technical specification, a product description or a new service design, how could the quality of that output be assessed? What methodology tells us whether an auto-innovation loop is good enough to act on?

Why It Matters

Level-5 autonomy is the point where the network no longer needs a human to approve its decisions. That only works if those decisions are demonstrably good and demonstrably safe. Technical trust, not organizational sign-off, is what unlocks the next era of connected intelligence. We’re looking for scalable, verifiable ideas that can be tested in real networks and applied widely.

Quantum Technologies: From Quantum Resources to Network Services

Imagine a network where quantum resources can be accessed as easily as connectivity or cloud compute today. Entanglement, quantum-state transmission and other quantum capabilities could become dynamically available across users and locations — turning emerging quantum infrastructure into services that can be discovered, orchestrated and consumed on demand.

We’re looking for bold technical innovations that bring quantum technologies into real-world network environments: from architectures for distributing and orchestrating quantum resources to novel applications, integration with classical telecom and cloud infrastructure, and service models that can turn quantum capabilities into tangible customer value.

What We’re Looking For

Here are some priority areas, though we welcome bold, creative ideas beyond this list:

• How can quantum resources become a network service? How could the transmission of quantum states or the distribution of entanglement be offered on demand, between different users and locations, much like connectivity and cloud resources are provided today?
• What are the killer applications for networked quantum resources? Beyond established concepts such as QKD, which novel applications could create tangible value from remotely shared entanglement or the ability to transmit qubits? Which customer segments and use cases would pay for quantum-network services beyond QKD, and what problem would quantum solve better than classical alternatives?
• How should a quantum network be controlled and orchestrated? What new architectures, protocols and control mechanisms are needed to discover, allocate, route and assure quantum resources dynamically across heterogeneous network infrastructures? What interim architectures can integrate quantum capabilities with existing fiber, optical transport, cloud and security infrastructure before mature end-to-end quantum networks exist?
• How can quantum and classical networks work together? How can quantum communication capabilities be integrated with existing telecom and cloud infrastructure, including classical optical networks, compute resources and applications, while remaining scalable and operationally viable? What operational capabilities would be required to assure, monitor and troubleshoot quantum services in a carrier environment with classical network operations?
• How can quantum-network services become commercially viable? What service models, performance metrics, SLAs and business models could turn scarce and technically demanding quantum resources into differentiated services that customers are willing to use and pay for? How should operators prioritize quantum investments relative to nearer-term AI-native network, edge, security and resilience capabilities?

Why It Matters

Quantum networking could extend telecommunications beyond the transmission of classical information and introduce entirely new network resources and capabilities. The opportunity is not only to build quantum links, but to make quantum resources accessible, orchestratable and useful as part of future network and cloud infrastructures.

We’re looking for scalable ideas that bridge the gap between quantum technology and real-world telecom networks — and help identify where networked quantum resources can create value that classical networks cannot.

Differentiated Services: The Next Evolution of Network Slicing and Intent-Based Networking

Imagine a network where differentiation isn’t a static, pre-configured slice policy, but a continuously negotiated outcome between what a service asks for and what the network can autonomously deliver. Today’s differentiated connectivity, built on 5G SA slicing, relies on largely static policies that become difficult to scale as more concurrent services compete for the same resources and policies begin to conflict. Intent-based networking offers a path beyond this: instead of pre-defining rigid slice configurations, services express what they need, and the network autonomously interprets, allocates and continuously assures it. This shift also opens a path toward more verifiable, enforceable SLAs, and, extended into the RAN, toward radio infrastructure that understands the services it’s carrying, not just the traffic it’s moving.

What We’re Looking For

Here are some priority areas, though we welcome bold, creative ideas beyond this list:

• Intent-Based Differentiation: How could networks move beyond static slice policies toward intent-based frameworks that autonomously interpret service requirements and resolve conflicts between concurrent regular, premium and AI-driven services in real time?
• SLA-Grade Differentiated Services: How could intent-based connectivity be turned into commercially viable, verifiable SLAs, enabling operators to sell guaranteed performance for premium or mission-critical use cases rather than best-effort differentiation?
• Self-Aware RAN: How could the RAN become aware of the underlying services it’s carrying, detecting and differentiating service types with less dependence on OS- or device-level signaling, to make smarter, service-informed resource-allocation decisions?
• Intent Assurance at Scale: How could a network continuously monitor whether it’s meeting a service’s intent, and autonomously act to keep performance within target limits, across many concurrent and possibly conflicting intents?
• AI-Enabled Resource Allocation: How could AI be used to holistically allocate radio resources across concurrent services, and to translate raw network telemetry into the higher-level performance indicators an intent framework needs?

Why It Matters

Static slicing got operators to differentiated connectivity, but it doesn’t scale as demand and use cases multiply. Intent-based networking, extended all the way into the RAN, offers a foundation for services to express what they need and have the network autonomously deliver and assure it, opening the door to SLA-backed premium services and to a RAN that understands, not just carries, the services running over it. We’re looking for scalable, deployable ideas that can turn this shift from a slicing evolution into a genuine new source of differentiated, monetizable network value.

6G Areas of Interest

Intent-Based Networks

Imagine telling the network what you need, rather than specifying how it should be configured. Instead of selecting from predefined products or manually configuring network parameters, users, businesses, applications, and AI agents express the outcome they need. The network then translates that intent into the required services and behavior.

Intent-based networks create the opportunity to make network services more dynamic and personalized. An intent could come from a person, an application, a machine, or an AI agent, with the network interpreting the request, configuring the necessary resources, and continuously adapting as needs change.

What We’re Looking For

Here are some priority areas, though we welcome bold, creative ideas beyond this list:

• Intent capture: How might a device understand what someone wants, spoken, gestured, or sent by an AI agent, and hand it straight to the network?
• Intent translation: How could a stated wish become a running service, with no engineer anywhere in the loop?
• Networks on demand: How might a trusted, personal network environment appear for exactly as long as it is needed, then disappear?
• New services & monetization: How could intent-based networks enable entirely new services and business models, turning user or machine intent directly into value?
• Machines with intent: How could a network serve millions of robots and AI agents asking for things at machine speed?
• Intent from others: How could a streaming service, a game studio, or a factory platform buy the network behavior it needs, safely and at arm’s length?
• Personal without preferential: How can one customer receive a service shaped entirely around them, while the network stays inside net neutrality rules?
• When intent goes wrong: How can a single misunderstood intent be prevented from reconfiguring the entire network, and quickly reversed before it causes disruption?
• Into the real world: How can these solutions work reliably in real, multi-vendor networks that were not designed to support them?
• Context-awareness: How can these solutions become context-aware to make decisions based on the status of the customer and/or terminal to optimize customer experience and resource efficiency?
• Context/intent memory: How to implement a context/intent memory in a regulation-compliant, secure and transparent way such that the customer is always in full control of their data?
• Service Awareness: How can we improve customer experience by implicitly deriving optimization measures based on the awareness of the services that are being used?

Why It Matters

A network that responds to intent no longer sells the same product to everyone. Instead, it delivers personalized experiences for individuals, provides businesses with measurable outcomes, and produces only what is specifically requested. We are seeking ideas that can be tested, validated, and widely applied.

Sensing

Imagine a mobile network that doesn’t just connect the world — it can sense, understand and respond to it. Integrated sensing could turn networks into a source of real-time intelligence, creating new possibilities for Physical AI, automation and smarter services.

We’re looking for bold solutions that make network sensing scalable, trustworthy, privacy-preserving and actionable.

What We’re Looking For

Here are some priority areas, though we welcome bold, creative ideas beyond this list:

• Sense, Understand & Act: How could mobile networks use integrated sensing to understand what is happening in the physical environment and autonomously respond to events, changes or risks?
• Sensing as a Network Service: How could operators turn network sensing into a scalable service, exposing useful insights about objects, movement and the environment to applications through standardized and easy-to-use APIs? How to enable intelligence on top of the sensing data to provide customers with answers to their questions, not just data about objects. How to know in “real-time” if we can meet the sensing request requirements from the customer. The Ray tracing approach is quite complex and compute intensive, on the other hand abstraction can be too far from the reality.
• ISAC for Physical AI: How could network-based sensing, connectivity and edge intelligence provide robots, drones, vehicles and other Physical AI systems with reliable situational awareness beyond what their onboard sensors can provide?
• Privacy-Preserving Sensing: How could network sensing deliver valuable information while minimizing collection and exposure of information about people and ensuring privacy, security, transparency and appropriate control over sensing data? The privacy is a high priority for us.
• Autonomous Communication–Sensing Orchestration: How could AI dynamically balance communication and sensing resources—including spectrum, beams, power and compute—to deliver sensing capabilities without compromising communication services or energy efficiency?
• Cooperative and Multi-Source Sensing: How could sensing information from multiple base stations, devices, vendors or sensing technologies be combined to create more reliable and accurate awareness of the physical environment? How to manage the presence or lack of trust in the data to provide trustworthy service?

Why It Matters

Network sensing could transform mobile networks from communication platforms into intelligent systems that understand the physical world. By combining sensing, connectivity and AI, networks can enable safer automation, smarter services and new capabilities for Physical AI. We’re looking for scalable, trustworthy and privacy-preserving ideas that can turn sensing into real-world value.

Network & Compute for Physical AI

Physical AI brings intelligence into the physical world. Robots, drones, vehicles, and other intelligent machines continuously reason based on what they sense in their environment, they act based on it adapt and learn. These functions have very different requirements and do not need to happen in the same place—some run directly on the device, while others can benefit from compute and intelligence distributed across the network edge and cloud.

6G creates the opportunity to build a network and compute continuum for Physical AI, connecting devices, radio sites, edge, and cloud. By combining communication, sensing, and distributed compute, the network can provide AI workloads with the resources, performance, and reliability they need—where and when they need them.

What We’re Looking For

Here are some priority areas, though we welcome bold, creative ideas beyond this list:

• Predictable performance: How can network and compute infrastructure provide the latency, reliability, and resilience that Physical AI applications need in real-world environments?
• Intelligence on the move: How can AI workloads seamlessly follow moving robots, vehicles, or drones across cells and sites?
• AI at the network edge: How can demanding AI workloads run within the power, space, and cost constraints of distributed network infrastructure?
• Communication-compute coordination: How can communication and computing resources be jointly coordinated to dynamically place and execute Physical AI workloads across device, edge, and cloud?
• Coordinating machines at scale: How can network and compute infrastructure enable large fleets of robots and autonomous machines to operate and coordinate efficiently?
• Sovereign AI infrastructure: How can Physical AI workloads run across distributed network and compute infrastructure while meeting European requirements for data and operational sovereignty?

Why It Matters

We want solutions that demonstrate how network and compute infrastructure can extend the capabilities of Physical AI beyond the device. The goal is to combine connectivity, sensing, and distributed compute in ways that enable Physical AI workloads to run reliably, efficiently, and at scale across real network infrastructure.

Beyond 6G

AI-Native and Autonomous Networks

Today’s networks are complex systems that require continuous configuration, optimization, and troubleshooting across many domains and vendors. As this complexity grows, manually operating every part of the network will no longer scale.

AI-native 6G creates the opportunity for a network that can understand its own state, anticipate what is needed, and increasingly manage itself. AI could optimize coverage and capacity, reduce energy consumption, assure services, identify the root cause of problems, and coordinate actions across RAN, core, and other network domains—with humans remaining in control of objectives and critical decisions.

Key Questions We Want to Explore

• AI-native network functions: How can AI be natively integrated into network functions across RAN, core, and other domains to continuously improve performance, coverage, capacity, and energy efficiency?
• From detection to resolution: How can AI identify anomalies, determine their root cause, and take corrective action before they affect services?
• Autonomy across the network: How can AI coordinate decisions across RAN, core, transport, and cloud rather than optimizing individual domains in isolation?
• AI agents for network operations: How can autonomous agents plan and execute complex network tasks while interacting safely with existing systems and other agents?
• Trust and control: How can autonomous network decisions remain explainable, auditable, and under operator control?
• Network for AI: How can networks provide the connectivity, performance, and data capabilities needed to efficiently support distributed AI workloads?
• Learning from the network: How can network data and digital twins be used to train, test, and continuously improve AI-driven network operations?
• Autonomy in a multi-vendor world: How can AI-driven automation work reliably across equipment, software, and interfaces from different vendors?

Why It Matters

We want solutions that move network automation beyond predefined rules and isolated AI use cases toward networks that can increasingly understand, optimize, and operate themselves. The goal is to make autonomous network operations reliable, controllable, and applicable across real, multi-vendor network environments.