AI agents expose 84 5G core security risks in telecom networks

The discovery of these 5G core security risks signals that the traditional 'walled garden' approach to mobile infrastructure is no longer viable in cloud-native deployments. As streaming providers increasingly rely on 5G MEC (Multi-access Edge Computing) for low-latency delivery, these vulnerabilities in the User Plane Function could expose high-value media traffic to interception or redirection. This shift forces a move toward zero-trust architectures where internal signaling requires the same rigorous authentication as external traffic. Watch for telecom vendors to accelerate patch cycles for PFCP and GTP-C interfaces as these AI-driven discovery methods become standard tools for both researchers and adversaries.

The discovery of 84 vulnerabilities in 5G core networks arrives as telecom vendors race to deploy agentic AI across their own network operations stacks. In June 2026, Ericsson launched its AI in RAN commercial software subscription, claiming up to 20% higher downlink throughput across more than 15 live deployments, while Nokia announced a GPU-accelerated AI-RAN partnership with Indosat Ooredoo Hutchison in Indonesia, expanding an architecture already adopted by T-Mobile US, SoftBank, and Vodafone. The irony is stark: the same AI-driven automation that promises operational efficiency also expands the attack surface that researchers at Nanyang Technological University have now demonstrated can be systematically probed by multi-agent systems like iFinder.

Nokia has moved aggressively to position its Autonomous Network Fabric as the control layer for this new era. At DTW Ignite in June 2026, Nokia announced partnerships with AWS and Databricks to build unified data, cloud, and control layers for autonomous networks, claiming operators are already achieving automation rates above 90% and service delivery times of four hours or fewer. Meanwhile, Ericsson adopted an agentic AI blueprint defining a service experience layer spanning customer journeys, revenue management, and network operations, running on Amazon Bedrock with more than 20 cloud-native AI applications positioned across OSS and BSS functions. These deployments underscore the urgency of the security findings: as vendors embed AI agents deeper into network control planes, the internal signaling interfaces that iFinder exposed become even more critical attack vectors.

The competitive divergence between Ericsson and Nokia on AI-RAN architecture has direct implications for how operators will address the vulnerabilities identified in this study. Ericsson and Nokia are now pursuing fundamentally different approaches to AI-RAN, with Nokia building its entire Layer 1 RAN strategy on Nvidia's CUDA platform and GPUs, while Ericsson focuses on software-layer AI optimization atop existing baseband silicon. Verizon disclosed that its 60,000-site vRAN is applying agentic AI to configuration changes and network optimization, while publicly calling for industry-wide interoperability standards for agentic systems. That call for standardization now carries additional weight: without agreed-upon security frameworks for , the class of vulnerabilities that iFinder uncovered in PFCP and GTP-C interfaces could multiply as each vendor's agentic stack introduces its own proprietary signaling patterns across cloud-native 5G core deployments.