When Connectivity Is Not Enough: Cross-Layer Attacks on UAV C2 over 5G

arXiv:2603.04662 · cs.CR · Submitted 2026-03-04 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.

Nadia: We're looking at the paper: "When Connectivity Is Not Enough".

Elias: Here's the gist of the paper. Beyond Visual Line of Sight (BVLOS) unmanned aerial vehicle (UAV) operations increasingly use 5G standalone (SA) networks for command and control (C2) between the UAV and the ground control station (GCS).

Nadia: Connectivity as Insufficient Security Indicator Operators may treat registration state, Protocol Data Unit (PDU) session status, and IP reachability as evidence that the C2 path is available.

Elias: In practice, these connectivity indicators alone do not guarantee that closed-loop control remains operationally safe.

Nadia: Attacks can degrade UAS C2 when timeliness degrades under shared User Plane contention, mobility continuity fails during Control Plane instability, or command integrity is violated at a trusted next-generation Node B (gNodeB).

Paper summary: Nadia: So to recap, this paper by Wagner Comin Sonaglio et al. is focused on Beyond Visual Line of Sight UAV operations that use 5G Standalone networks for command and control between a drone and its ground station <ref:2603.04662#pg1,Beyond Visual Line of Sight>. The central thesis is that treating connectivity indicators like registration state or IP reachability as proof of a secure C2 path is insufficient for safe operation.

Elias: They build this reproducible testbed using Open5GS and UERANSIM, running MAVLink over the 5G User Plane, grounded in assumptions from a Nokia CMU 5G SA core <ref:2603.04662#pg1>. This setup allows them to simulate real-world traffic placement and reachability scenarios accurately.

Priya: The paper claims that C2 can degrade when timeliness suffers under shared User Plane contention, or when mobility continuity fails during Control Plane instability, or even if command integrity is violated at a trusted next generation Node B after PDCP deciphering.

Nadia: They specifically define three threat models to cover these scenarios: one stressing shared queueing on the User Plane, another stressing service-based interfaces by targeting reachability during mobility, and a third focusing on the RAN trust boundary where they rewrite cleartext payloads after radio layer deciphering.

Elias: The overall claim is that connectivity-centric metrics can misread risk for cyber-physical C2 over 5G SA <ref:2603.04662#pg1>. They show that things like stale telemetry or command hijacking can occur while the GTP User tunnel remains active, which is a key finding in this work.

Priya: It matters because it shifts the focus from just network availability to the actual safety of the physical control loop. It means we need to look at bounded delay and coherent telemetry, not just whether a session is established.

Nadia: Exactly. They argue that for BVLOS operations, connectivity alone doesn't guarantee closed-loop safety; you need to evaluate the control loop itself against adversarial contention and instability.

Conclusion: Elias: Looking at the title, "When Connectivity Is Not Enough," it really hits home that we often mistake a network link for control assurance. The authors are showing us that in complex systems like UAV command and control over 5G SA, the link is just one piece of a much larger puzzle <ref:2603.04662#pg1>.

Nadia: I think what this paper does is make sure operators stop relying solely on the IP address being reachable as their security signal for drones. It pushes them to look at how commands are timed and whether those commands are actually what they think they are, even when the tunnel is technically up.

Priya: So the practical implication here isn't just a technical fix, it’s a mindset change for deploying these systems. It means designing for failure in the control loop itself rather than assuming network layers handle all the security guarantees.

Elias: That’s right. The paper demonstrates that you can have connectivity metrics looking nominal while critical aspects like command integrity or timeliness fail because of contention or instability, and that's where the real risk lies.

Nadia: Ultimately, this research grounds the discussion in cross-layer failures—timeliness, availability during movement, and RAN compromise—showing that safe UAV operations depend on evaluating the control loop under those specific adversarial conditions.

Elias: It’s about moving beyond simple reachability checks to a deeper analysis of how the entire communication stack interacts when an adversary is present.

Aeronautics Institute of Technology · Northeastern University

cs.CR

Submitted: 2026-03-04

Updated: 2026-11-12

Code: https://github.com/0wln3d/connectivity-not-enough-artifact

Project page: https://gradiant.github.io/open5gs-operator

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 86/100

The gist: Beyond Visual Line of Sight (BVLOS) unmanned aerial vehicle (UAV) operations increasingly use 5G standalone (SA) networks for command and control (C2) between the UAV and the ground control station…

Key concepts

Connectivity as Insufficient Security Indicator
Operators often assume a UAV's C2 link is secure just because it shows a successful registration state or IP reachability. The paper demonstrates this is dangerous because connectivity doesn't guarantee safe control; attacks can still interfere with timely command delivery or stability.
Post-Attachment Threat Models
The study used three specific scenarios to test security risks: one focusing on shared network contention (TM1), another targeting control plane access during movement (TM2), and a third exploiting a compromised base station's radio layer (TM3). These models stress different parts of the 5G system.
Cross-Layer Failure Mapping
The research maps specific threats to the properties they impact: TM1 stresses timeliness, TM2 stresses availability, and TM3 shows semantic integrity failure. The main takeaway is that connectivity loss isn't the problem; it's the loss of control guarantees once a connection exists.
Control Loop Evaluation
Safe UAV operations must be evaluated by testing the entire control loop against adversarial conditions like contention or compromise, rather than just checking if a connection is established. The paper emphasizes evaluating how the system behaves under these real-world stresses.

Terminology

Summary

Beyond Visual Line of Sight (BVLOS) unmanned aerial vehicle (UAV) operations increasingly use 5G standalone (SA) networks for command and control (C2) between the UAV and the ground control station (GCS). p. 0

Connectivity as Insufficient Security Indicator

Operators may treat registration state, Protocol Data Unit (PDU) session status, and IP reachability as evidence that the C2 path is available. In practice, these connectivity indicators alone do not guarantee that closed-loop control remains operationally safe. p. 0 Attacks can degrade UAS C2 when timeliness degrades under shared User Plane contention, mobility continuity fails during Control Plane instability, or command integrity is violated at a trusted next-generation Node B (gNodeB). p. 0 Such failures undermine connectivity as the central security indicator for UAV operations. p. 0

Post-Attachment Threat Models

The paper demonstrates three distinct threat models on a reproducible Open5GS and UERANSIM testbed that carries Micro Air Vehicle Link (MAVLink) over the 5G User Plane, and it uses a commercial Nokia core to ground deployment assumptions. These threat models stress three interfaces: TM1 stresses shared queueing on the User Plane, TM2 stresses service-based interfaces by sending black-box HTTP/2 requests to the Network Repository Function (NRF) and the Session Management Function (SMF) during mobility, and TM3 stresses the radio access network (RAN) trust boundary by rewriting cleartext payloads on N3 after radio-layer deciphering on a compromised gNodeB. p. 0

Threat Model Analysis

The three threat models are defined as:

  1. TM1 represents a threat model focused on the degradation of C2 timeliness, where an adversary is a co-tenant UE assigned to the same slice and DNN as the UAV and/or GCS. p. 3

  2. TM2 addresses an insider adversary with access to the Control Plane, targeting Control Plane availability during mobility through SBI reachability. p. 5

  3. TM3 focuses on a compromised gNodeB where the adversary has full control of the RAN node, leading to command tampering after PDCP deciphering. p. 3

Mitigation and Conclusion

Mitigations evaluated include slice/DNN isolation, forwarding filtering, SBI segmentation with mTLS, and MAVLink signing. The results show that connectivity-centric metrics can misread risk for cyber-physical C2 over 5G SA. p. 1 Safe UAV operations require evaluating the control loop itself under adversarial contention, Control Plane instability, and RAN compromise, rather than treating registration and IP reachability as sufficient evidence of security. p. 1

Cross-Layer Failure Mapping

Table IV maps each threat to the property it stresses: TM1 stresses Timeliness, TM2 stresses Availability, and TM3 shows that the semantic integrity of MAVLink commands can fail with active tunnels because the trusted RAN path can still expose application octets in cleartext to a compromised forwarder. The common failure is not the absence of connectivity but the loss of control guarantees once connectivity is established. p. 0

Ethical Considerations

All experiments in this work were performed in controlled testbeds operated by us. We did not run attacks against public mobile networks, third-party infrastructure, production UAV systems, or equipment outside our authorization. The Nokia CMU 5G SA setup was used only to ground reachability and traffic-placement assumptions in a commercial core; we did not attempt to exploit vulnerabilities in that platform. p. 1

Related Work Context

Our work uses the UAV C2 loop as the common observation point from a cross-layer perspective. The contribution of this paper is therefore not reducible to one protocol, one CVE, or one interface. p. 6

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Enhanced security posture by moving beyond connectivity metrics to evaluate control loop safety, as connectivity is not safe control and requires testing under adversarial contention, Control Plane instability, and RAN compromise.

  2. Improved resilience against User Plane contention by implementing slice/DNN isolation and forwarding filtering while explicitly recognizing that filtering near the UPF is necessary but not sufficient alone.

  3. Strengthened Control Plane availability during mobility by deploying SBI segmentation with mutual TLS to reduce insider reachability to NRF and SMF, addressing the failure where Control Plane instability prevents session recovery during handover.

  4. Increased semantic command integrity by mandating MAVLink signing on the UAV and GCS to block tampered commands, as MAVLink signing also introduced two practical constraints and serves as a practical mitigation for TM3.

  5. Improved operational awareness by incorporating robust state validation beyond schema checks within NFs to prevent fatal invalidstate failures caused by crafted requests, addressing the issue where Schema validity does not guarantee semantic validity with respect to the NF runtime state.

Sources

Related papers