2027 IEEE International Conference on Communications (ICC) - Hidden Costs of AI-Native Next-Gen Networks: Workshop on Trade-Offs in Sovereignty, Security, reSilience, and Sustainability (4S-Net)

Scope of the Workshop

Networks are becoming a key enabler of the confluence of communications and AI faster than the community can assess the consequences. Learning-based control, management, and security are spreading across the next-generation stack, from 5G-Advanced/6G and optical/IP transport to AI-factory fabrics, the cloud–edge continuum, non-terrestrial, and industrial networks. Yet deployments are often assessed by the objective they improve, while costs elsewhere receive less attention. In AI-native networks, sovereignty, security, resilience, and sustainability are increasingly coupled: continuous anomaly detection adds inference load, sovereign model hosting can reduce hyperscale efficiency and vendor diversity, and redundancy and intelligent orchestration can introduce new failure modes. These trade-offs remain largely unquantified. The workshop seeks to make these tensions explicit, measurable, and negotiable before they become embedded in infrastructure. It addresses directions of the communications–AI convergence: AI in the network, through learning-based control, management, and security, and the network for AI, through infrastructure supporting AI workloads. Both call for a shared methodology including Key Value Indicators alongside conventional KPIs for assessing multi-objective trade-offs rather than isolated gains. The topic is particularly timely for ICC 2027, whose theme, “Confluence of Communications and AI,” reflects this transition. AI in networks is moving towards broader adoption, while mobile networks progress towards initial normative 6G specifications. Meanwhile, regulatory, security, data-localisation, and energy constraints increasingly differ across jurisdictions, influencing where infrastructure and models can be deployed. Sovereignty is therefore becoming an engineering constraint alongside efficiency, resilience, and security. Decisions made now—where inference is placed, who controls the models, and which network functions become learning-based—can shape infrastructure well into the 2030s. The workshop provides a forum to examine these costs and trade-offs while the design space remains open, complementing recent ICC/GLOBECOM workshops on generative AI, foundation models, AI-native RAN, trust, and green networking, which have largely focused on individual dimensions. The workshop explores, in particular, the following trade-offs: 

1) Sovereignty vs. AI Innovation: Sovereign control over infrastructure, data, and model weights can constrain vendor choice and innovation, while reliance on global AI platforms can create cross-border data flows, legal exposure, and model-layer lock-in. Topics include sovereign telecom models, model/data provenance, confidential computing, sovereign cloud/edge deployments, and AI supply-chain risks.

2) Security, Resilience vs. Sustainability: AI-based detection, zero-trust enforcement, inline inspection, post-quantum cryptography, and distributed redundancy can improve security and resilience while increasing compute, memory, energy, and hardware costs. Conversely, energy-efficient and geographically concentrated infrastructure can conflict with data localisation and redundancy. We seek work that quantifies these trade-offs, including energy and carbon costs per prevented incident, the overhead of resilience mechanisms, and points of diminishing returns.

3) Resilience vs. AI-induced complexity and autonomy: Multi-path networking, distributed control, and digital twins can improve availability while making AI a critical dependency. Model drift, hallucinated configurations, adversarial inputs, foundation-model monoculture, and agentic-AI risks such as prompt injection and unsafe actions introduce new failure modes, motivating requirements for guardrails, authorisation, and auditability.

4) Measuring the Trade-offs: We invite multi-objective and Pareto optimisation, KVI frameworks and benchmarks, measurement studies of AI-native overheads, explainable AI for accountable operation, and digital twins or testbeds for pre-deployment evaluation. Measurement studies, negative results, and position papers are welcome. Keynotes and a panel will complement the paper sessions, aiming to develop a shared vocabulary for the costs and trade-offs of AI-native networks across wired and wireless, terrestrial and non-terrestrial, carrier, enterprise, and industrial settings.

 

Suggested topics

  • Security, resilience, and sustainability, including Pareto-based approaches and KVI frameworks
  • Sovereign and globalised AI for networks, including foundation models, model/data provenance, ownership, and jurisdictional and data-locality constraints
  • Security and sustainability trade-offs, including the energy and carbon costs of AI-based detection, zero-trust, inline inspection, and post-quantum cryptography
  • Security and resilience economics, including cost per prevented incident, redundancy overhead, and points of diminishing returns
  • Reliable and resilient AI-native networks, including model drift, adversarial inputs, foundation-model monoculture, multi-connectivity, satellite–terrestrial integration, and distributed control
  • Agentic AI in network operations, including attack surfaces, guardrails, authorisation, auditability, and safe autonomous control
  • Networks for AI, including transport, data-centre, and edge infrastructure under energy, grid, sovereignty, and resilience constraints
  • Accountable AI and pre-deployment evaluation, including explainable AI, digital twins, testbeds, deployability assessment and measurement methodologies

 

Important Dates:

Papers submission deadline: 17 January 2027
Paper Acceptance Notification: 07 March 2027
Camera Ready: 21 March 2027

 

Workshop Co-Chairs

Frank Loh, University of Würzburg
Elisa Rojas, Universidad de Alcalá
Christoph Lipps, German Research Center for Artificial Intelligence
Lena Wosinska, Chalmers University of Technology
Yulei Wu, University of Bristol
Hesham ElBakoury, Hitachi
Marco Savi, University of Milano-Bicocca