You Can't Patch the Silicon: Post-Quantum Migration Realities in Automotive HSM and AUTOSAR Crypto Stacks

Post-quantum migration guidance is largely written for systems that can be patched in place: swap a TLS library, rotate a certificate authority, roll out a firmware update to a fleet of servers. Automotive embedded systems break almost every one of those assumptions, and this talk is a practitioner's account of what actually happens when you run a real PQC risk and migration assessment against an in-vehicle HSM and AUTOSAR crypto stack instead of a data center.

Three constraints define the problem. First, SecOC authenticates every CAN and CAN-FD frame in real time, inside payloads that top out around 64 bytes — ML-DSA signatures and ML-KEM ciphertexts run into the kilobytes, so the real-time authentication path cannot simply be re-signed with a post-quantum primitive; the migration has to happen somewhere else in the chain. Second, the HSM is fixed-function silicon soldered into a vehicle that will still be on the road in fifteen to twenty years, with no OTA path to add lattice-math acceleration after the fact — so "harvest now, decrypt later" isn't an abstract future risk, it's a clock already running against hardware that shipped years ago. Third, the AUTOSAR crypto stack (CSM, CryIf, KeyM) was built around parameter assumptions that classical algorithms never forced anyone to question, which turns crypto-agility into a real integration problem rather than a configuration flag.

This talk walks through where post-quantum primitives map cleanly today (OTA backend TLS, V2X certificate issuance, secure boot signing), what has to stay symmetric and why, and how an ISO 21434-aligned risk assessment was used to decide which ECUs and assets get migrated first when neither silicon nor budget allow migrating everything at once.

Key Takeaways

Attendees will leave with:

• A concrete picture of what does and doesn't map onto NIST's PQC primitives in a real-time embedded authentication path, and why
• A working model for why "harvest now, decrypt later" timelines look different for fixed-lifecycle hardware than for patchable infrastructure
• A risk-based method, grounded in ISO 21434 clause 8, for sequencing PQC migration across an ECU portfolio under real budget and silicon constraints
• Open questions the automotive supply chain hasn't answered yet — useful for anyone assessing vendor or supplier readiness in this sector