AI Agents Expose Critical 5G Core Vulnerabilities: Why Internal Signaling and Cloud Security Require Zero Trust

A recent report by Fierce Network highlights a major security finding: researchers using an autonomous multi-agent LLM system (iFinder) uncovered 84 previously unknown vulnerabilities across seven commercial 4G and 5G core implementations, 81 of which received official CVEs.

The discovery underscores a major paradigm shift—AI is accelerating the speed of vulnerability discovery far faster than traditional operator patch cycles, proving that internal network interfaces can no longer be treated as secure or trusted environments.

In today’s hyper-connected networks, operators face several critical security exposures:

• Internal Interfaces as Active Attack Surfaces: Interfaces like N4/PFCP in 5G and S11/S5/GTP-C in 4G were historically assumed to be secure because they resided inside the telecom perimeter. AI agents demonstrated that these internal paths are highly vulnerable, making mutual authentication and deep message validation mandatory.

• Subscriber Traffic Hijacking: Among the findings was CVE-2026-8233, a User Plane Function (UPF) session hijacking vulnerability. Validated on live commercial 5G networks, this flaw allows an attacker with access to an internal interface to redirect subscriber uplink traffic, threatening data privacy and enterprise services.

• Automated Threat Chaining: AI agents excel at stringing together minor flaws across multiple technologies, building sophisticated attack chains that bypass conventional perimeter defenses.

Addressing these vulnerabilities requires a comprehensive, multi-layer approach:

• Multi-Protocol Firewalling: Operators must deploy cross-generational signaling protection spanning SS7, Diameter, HTTP/2, and GTP/PFCP protocols. This prevents attackers from exploiting legacy 2G/3G/4G entry points to compromise 5G core functions.

• Zero Trust for Service-Based Architectures: In cloud-native 5G environments, core interfaces require automated message validation, state-invariant checks, and strict parameter screening on HTTP/2 and N4/PFCP interfaces to block spoofed or hijacked API calls.

• Shielding Hybrid Cloud Deployments: Independent control-plane security platforms must continuously monitor and inspect signaling traffic across bare-metal, virtualized, and public/private cloud environments to shield network functions from internal and external threats.