eSIM security vulnerabilities

See Security Explorations - eSIM security (by Adam Gowdiak) for what I consider the most comprehensive published applied research into eUICCs.

I’ve shared some of my initial thoughts on this in a blog post at Security Issues regarding GSMA eSIMs / eUICCs + Javacard | LaForge's home page

4 Likes

With eSIM, why the telco industry didn’t move to TEE, TrustZone, TPM, etc? That has hardware isolation instead of “applet firewall” and “personalization” and other 1990s sentimentality. Even cheap micros have this stuff now.

Because that’s all been broken again and again and again.
Intel SGX: Full AACSess: Exposing and exploiting AACSv2 UHD DRM for your viewing pleasure - media.ccc.de

TEE: Ramming Enclave Gates: A Systematic Vulnerability Assessment of TEE Shielding Runtimes - media.ccc.de

TrustZone: TrustZone-M(eh): Breaking ARMv8-M's security - media.ccc.de

These “security” features are not a good way to actually protect sensitive information and the industry really wants to protect the information inside the SIM.

2 Likes

I see your point, but as per Adam Gowdiak, it seems like eSIM + Javacard may have its own weakness.

@mayhaze Adam is overgeneralizing his assumptions based on the results of this implementation. In fact, I predicted problems with this implementation well over a year in advance, as it was an ECC software implementation lacking the benefits of strong hardware key protection and a dedicated crypto coprocessor.

“Mature” products from companies like G&D, THALES DIS (formerly Gemalto → Gemplus + Axalto → Schlumberger), NXP, and even Infineon, etc., are completely different. I work in this field and have found and reported several flaws in these systems over the past few decades.

@Manawyrm is right regarding TEE and TZ (SGX is irrelevant here): Every implementation is at least partially proprietary (the Secure Boot/High Assurance Boot for loading the Secure Supervisor differs). Some offer stronger security guarantees, while others shouldn’t be trusted. There have also been several studies on this topic by the industry and the BSI (German Federal Office for Information Security).

TL;DR: Even if the implementation, as seen in this example, is flawed, it’s still better to distribute secrets across different/dedicated chips to create a security barrier. While I don’t want to defend this implementation, it’s worth noting that exploiting the private ECC key required some prior privileges (loading code, accessing the physical device). As mentioned earlier, this wouldn’t have been possible in other products, even with failing virtual machine checks.

(I’m bound by NDAs, so I’ll deliberately be vague about specific details.)

1 Like