AI Hunt Exposes Telecom Core Trust Flaws

Nanyang Technological University researchers used AI agents to find 84 previously unknown flaws in 4G and 5G core software; 83 were confirmed, 81 now have CVEs, and 23 remain unpatched, including CVE-2026-8233, a session-hijack flaw in subscriber traffic handling. The attack works because core networks treat internal forwarding messages as trusted. An attacker who can reach that control link can send a higher-priority routing instruction that reuses an existing forwarding-rule ID, so the network picks the attacker’s rule and sends the subscriber’s outbound traffic to them instead. Researchers reproduced it in OpenAirInterface and in two commercial cores, and also showed it could ride inside a valid SIM data tunnel in some systems. The exposure sits in the trust model of carrier cores, not just in one codebase. If those interfaces are cloud-exposed or reachable through tunneling, the same design can turn a single forged internal message into live session interception across carrier-grade and derivative deployments.

Part of the PlainSec briefing for 2026-08-11

Every edition of this story: AI Hunt Exposes Telecom Core Trust Flaws

Sources