Foundation: 5G security is already strong
Start from what is solid. 5G introduced real improvements over earlier generations: mutual authentication (the network proves itself to the device, not just the reverse), and concealment of the permanent subscriber identity — the SUPI is encrypted into a SUCI before it ever crosses the air, defeating the classic "IMSI catcher". 6G inherits this architecture; the job is to extend it, not rebuild it.
5G's security architecture (mutual auth, encrypted subscriber identity, integrity protection of signalling) is the baseline 6G starts from. These mechanisms are well specified and carry forward conceptually.
3gpp.org 2026-06-15| Feature | Mechanism | 5G spec reference | 6G evolution direction |
|---|---|---|---|
| Subscriber identity concealment | SUPI encrypted to SUCI using ECIES with home-network public key before transmission over the air | TS 33.501 §6.12 | candidate PQC-hardened key |
| Mutual authentication | 5G-AKA or EAP-AKA′ — both UE and network prove identity; eliminates passive IMSI-catcher attack | TS 33.501 §6.1 | candidate PQC key exchange |
| NAS signalling integrity | NAS messages integrity-protected from registration; mandatory from connection setup | TS 33.501 §5.3.1 | foundation carried forward |
| RRC signalling integrity | RRC integrity protection activated at RRC connection establishment | TS 33.501 §5.3.4 | foundation carried forward |
| User-plane confidentiality | Encryption of UP traffic; operator-optional in 5G (AES-128/256, SNOW 3G, ZUC) | TS 33.501 §5.3.5 | candidate mandatory direction for 6G |
| Security algorithm negotiation | UE sends security capabilities; gNB/AMF select algorithm; bidding-down protected | TS 33.501 §6.7 | candidate add PQC algorithm classes |
Why the quantum era changes the maths
Today's key exchange relies on problems (factoring, discrete logs) that a large quantum computer could break with Shor's algorithm. The threat is not only future: an attacker can harvest now, decrypt later — record encrypted traffic today and decrypt it once a quantum computer exists. For a network whose data must stay secret for a decade-plus, that makes post-quantum cryptography (PQC) a design-time concern, not a later patch.
Moving key exchange and signatures to quantum-resistant algorithms (the kind standardised by NIST's PQC programme) is widely expected to shape 6G security, but the specific 6G mechanisms are not defined — this is direction-setting research, not a 3GPP specification yet.
itu.int — IMT-2030 2023-11| Algorithm | NIST standard | Family | Purpose | Replaces | 6G relevance |
|---|---|---|---|---|---|
| CRYSTALS-Kyber (ML-KEM) | FIPS 203 | Lattice (Module-LWE) | Key Encapsulation Mechanism | ECDH / RSA key exchange | Secures key agreement in mutual authentication (replaces ECDH in AKA) |
| CRYSTALS-Dilithium (ML-DSA) | FIPS 204 | Lattice (Module-LWE) | Digital signature | ECDSA / RSA signatures | Certificate and message signing in NF authentication |
| FALCON (FN-DSA) | FIPS 206 | Lattice (NTRU) | Compact digital signature | ECDSA (compact use cases) | Low-bandwidth signature for constrained devices / IoT UEs |
| SPHINCS+ (SLH-DSA) | FIPS 205 | Hash-based (stateless) | Conservative backup signature | RSA (high-assurance contexts) | Fallback where lattice assumptions are not trusted; larger signature size |
ITU-R M.2160 lists security and resilience among the 15 IMT-2030 capabilities — but as a qualitative one, with no numeric target. So it is a genuine requirement-level expectation that 6G be secure and resilient, without a specific number attached.
itu.int — IMT-2030 2023-11The AI-era threat model
6G's embrace of AI is double-edged. The same ML that optimises the network becomes an attack surface: poisoned training data, adversarial inputs that fool a model, and the privacy of the data models are trained on. An AI-native network has to defend its models, not just its traffic. This is new territory the security community is actively mapping.
Model integrity, adversarial robustness and training-data privacy for an AI-native air interface and core are recognised challenges without settled 6G answers. Flag the risk; do not imply it is solved.
itu.int — IMT-2030 2023-11Where this connects
Security spans the whole stack: it protects the AI functions, lives mostly in the core, and underwrites the trust that the requirements assume.