I'm in agreement, but then we get back to the fact that memory safety may not be a compelling argument to use a new language.
After all, most products aren't "worth attacking" until they are large and successful enough, thus reinforcing the decision not to rewrite a product in a new language. This means that memory safety alone is not a compelling argument for switching languages.
Until recently[1] payment terminals were memory constrained devices. Even right now, a significant portion of payment terminals are constrained (128MB of RAM, slow 32-bit processors).
Even though Java was around in 2006, the payment terminals I worked on then ran on 16-bit NEC processors (8088, basically).
Even if we aren't looking payment terminals (or any of the intermediaries), there's still a large and not-insignificant class of devices for which anything but C or a reduced set of C++ is possible. Having common libraries (zip, tls, etc) written in C means that those libraries are usable on all devices from all manufacturers.
[1] The industry-wide trend right now is a move towards android-based terminals. While this does mean that you can write applications in Java and Kotlin for these terminals, it's still cheaper to port the existing C or C++ codebase to Android and interface via JNI, as that reduces the costs (of which certification is a significant minority).
Even right now, there is more portability in writing the EMV transaction logic in plain C because then it can be used from almost anywhere. A team that went ahead and wrote the core payment acquisition logic in Java would find themselves offering a smaller variety of products and will soon get beaten in the market by those manufacturers who packaged the logic up into a C library.
Don't underestimate how price-sensitive embedded consumers are. A savings of a few dollars per product can absolutely lead to a really large leg up over the competition.