Implicit trust between network functions yields 84 flaws in 4G and 5G cores — 81 CVEs and session hijacking included

Late July 2026. July 27, 2026. July 31, 2026. Three dates that frame a discovery set to reshape mobile network security. 84 vulnerabilities — 81 already assigned CVE identifiers — have been uncovered inside the 4G and 5G core networks that handle authentication, mobility, and data routing for billions of subscribers. The culprit isn’t a one-off coding mistake. It’s a systemic design failure: network functions trust each other implicitly, executing any message a peer sends them without validation. And the shift to cloud-native deployments dismantled the only safeguard that ever protected them — physical isolation.

The findings come from a team at Singapore’s Nanyang Technological University (NTU), publishing on July 31, 2026 a paper titled Understanding Implicit Trust Errors in Core Carrier Networks through Multi-Agent Flaw Discovery and Analysis. The study spans seven open-source LTE and 5G core implementations — Open5GS, OpenAirInterface, free5GC, SD-Core, and eUPF — across two critical signaling protocols: GTP-C (GPRS Tunnelling Protocol Control Plane) and PFCP (Packet Forwarding Control Protocol). These aren’t academic toys: they power research testbeds, commercial deployments, and form the foundation of telecom vendor products.

iTrue: when the network core trusts with its eyes closed

The researchers named this vulnerability class iTrue, short for implicit trust errors. The concept is as simple as it is devastating.

Inside a mobile network, core functions — AMF (Access and Mobility Management Function), SMF (Session Management Function), UPF (User Plane Function), SGW-C (Serving Gateway Control Plane), PGW-C (PDN Gateway Control Plane) — exchange signaling messages to establish, maintain, and tear down user sessions. Every call, every data session, every SMS triggers a cascade of dialogues between these functions over GTP-C and PFCP.

Historically, these signaling interfaces sat behind physical isolation inside carrier data centers. The security perimeter was the building. No external threat could reach a GTP-C or PFCP message, so nobody ever bothered to validate their contents. A core function receives a message from its neighbor — it executes it. End of story.

Network Function Virtualization (NFV) changed everything. As 5G cores moved onto Kubernetes, OpenStack, and public cloud infrastructure, signaling interfaces migrated to shared overlay networks. One misconfiguration — a sloppy Network Policy, a BGP leak, a bridge between the user plane and control plane — and those interfaces become routable from the internet. The trust model, built for a world where physical perimeters were the security layer, turned fragile overnight. The 84 flaws prove it empirically.

“Our research finds these vulnerabilities share a single recurring root cause, implicit trust between core network functions, and are present in widely used open-source LTE/5G cores that back research testbeds and commercial deployments alike,” the authors write.

The attacks enabled by iTrue flaws fall into two categories:

• Denial of Service (DoS). An attacker sends a malformed GTPv2-C Create Session Request message to the SGW-C function. The function attempts to parse it without prior validation and crashes — session terminated for every subscriber attached to that instance. No authentication required, no special privileges: the interface just needs to be reachable.
• Session hijacking. This one is surgical. It exploits the absence of uniqueness checks on Packet Detection Rule identifiers (PDR ID) in PFCP. The scenario: the attacker establishes a legitimate PFCP association with the UPF, then waits for a victim to initiate a session. Once the SMF sends a PFCP Session Establishment Request for the victim, the attacker injects a PFCP Session Modification Request that reuses the victim’s PDR ID with higher priority and a malicious forwarding rule pointing to the attacker’s infrastructure. The UPF, which doesn’t verify PDR uniqueness, sorts rules by priority and places the malicious rule on top. The victim’s uplink traffic routes to the attacker, not the internet. The user notices nothing.

That last scenario is not theoretical. Two operators running commercial 5G cores were confirmed vulnerable to session hijacking. The first, Dotouch, has already patched the issue in XproUPF (CVE-2026-8233, CVSS 4.6). The second — an unnamed major 5G vendor — remains in remediation at the time of publication.

The attack can originate from a remote adversary (outside the core network, having obtained a core function’s IP address through public documentation, passive enumeration, or active scanning) or from a malicious User Equipment (UE) connected to the mobile network. In the latter case, the attacker smuggles crafted GTP-C or PFCP messages inside the uplink data stream (GTP-U), exploiting protocol tunneling to cross the boundary between user plane and control plane — a boundary many cloud-native deployments fail to enforce strictly.

iFinder: how AI systematically found what humans kept missing

The other half of the story is methodological, and it speaks directly to DevSecOps teams invested in automated vulnerability discovery.

The researchers didn’t manually audit seven complex codebases. They built iFinder, an LLM-assisted multi-agent system designed to discover implicit trust errors at scale. The pipeline runs in four phases:

1. Synthesis and categorization. The system ingests previously documented flaws in core networks and distills them into detection patterns — missing message format validation, missing semantic field validation, missing resource uniqueness checks.
2. LLM-driven discovery. Agents scan the source code of the seven target implementations (C, C++, Go), applying those patterns to identify potential iTrue candidates. At this stage, the false positive rate is high — the LLM “hallucinates” vulnerabilities that don’t exist.
3. Code-specification cross-checking. This is the key innovation. Each candidate is mapped to the protocol procedure it implements (from the GTP-C or PFCP specification), then the system verifies in the source code that the validation required by the specification is actually absent. This technique eliminates nearly all hallucinations.
4. PoC generation and execution. Surviving candidates are turned into proof-of-concept exploits, executed against real implementations in a controlled environment, and refined iteratively until confirmed or rejected.

The result: 84 previously unknown vulnerabilities, 83 confirmed, 81 CVE identifiers assigned to date. Some iTrue flaws discovered in 5G cores were inherited directly from their 4G counterparts — code was reused without anyone reassessing the trust assumptions in the new deployment context.

The iFinder framework illustrates a larger trend: LLM-driven automated vulnerability discovery, already demonstrated by Google on Chrome (1,072 flaws fixed across two releases), is reaching domains where human expertise is scarce and expensive — telecom signaling at the top of the list.

What this means for operators, cloud architects, and CISOs

For telecom operators. Verify three things immediately. One: network isolation of your GTP-C and PFCP interfaces — they must never be routable from the internet or the user plane. Two: micro-segmentation between core functions, with Network Policies enforcing least privilege. Three: message validation at every interface — a vendor patch alone doesn’t help if your architecture exposes those interfaces by default.

For cloud and DevOps teams. Migrating 5G network functions to Kubernetes is not a lift-and-shift. The security model that worked inside a private data center with physical isolation does not hold on a shared cluster. Pod-to-pod mTLS, strict Network Policies, and control-plane/data-plane separation are not nice-to-haves — they are security prerequisites. If your UPF and SMF share the same overlay network as your application workloads, you already have a problem.

For CISOs. The 81 CVEs cover open-source implementations, but the underlying issue — implicit trust — affects proprietary vendors too. The two affected commercial operators prove it. Add signaling interface validation to your telecom security audits. Demand that your core network vendors document their inter-function trust model and the validation mechanisms they apply on every interface.

Verdict

The 84 iTrue flaws are not another vulnerability bulletin — they document an entire class of structural weaknesses that spans mobile network generations, from 4G to 5G, from open source to proprietary.

If you operate a 4G or 5G core, your priority is not the latest vendor firmware patch. It’s re-segmenting your signaling interfaces so they return to what they should never have stopped being — unreachable from the outside. The 81 CVEs are the symptom. Implicit trust is the disease. And cloud-native, deployed without the right guardrails, is the accelerator.

References

• Understanding Implicit Trust Errors in Core Carrier Networks through Multi-Agent Flaw Discovery and Analysis, Ziyu Lin et al., Nanyang Technological University, July 31, 2026
• The Hacker News, Researchers Report 84 Flaws in 4G and 5G Cores, Including a Session Hijacking Flaw, July 31, 2026 — https://thehackernews.com/2026/07/researchers-report-84-flaws-in-4g-and.html
• CVE-2026-8233 — Dotouch XproUPF PFCP Session Hijacking
• CVE-2026-50522 — Open5GS LTE GTPv2-C Denial of Service