They wouldn't need to even talk to the AP to get out on to the network though :) If you have physical write access, you can talk to the network just by writing data in the network device's mailbox using the exactly same MMIO techniques that the AP uses. Its one of the reasons I am so disenchanted with current rootkit detection tech—if malware can get physical memory write, they can corrupt other cores memory persist into other SoC elements, well and truly out of view from the AP while still maintaining the exact same amount of control. I would not be surprised in the _least_ if the US three letter agencies have at least one implant that does this because its not exactly difficult but is so insanely hard to fix or detect that it makes it ideal for covert monitoring.

That being said, I think there's a fundamental disconnect between how SoCs work and how people actually think they work. Linux lives in just one tiny little corner of the machine and the hardware does so much to coddle Linux into thinking it's in control. The AP is just a core on the memory bus, it typically has no greater privileges (it often has _less_ because hardware vendors hide ""implementation details"" away from OS developers!) than any other core. I don't know that this is so much a radical take in the OS community though it might be one here: Linux _is_ part of the problem. Linux makes several assumptions about how hardware works. Chiefly, it assumes that it is in the driver's seat and it makes cooperating with other cores running their own copies of Linux (let alone other OSs) damn near impossible. I think the next step in computer evolution is going to come from software engineers rediscovering hardware and building code that not only runs on the AP but across all cores—from the power management controller to the AP--in an SoC. This may come from radically changing Linux or it may very well come from a new OS that treats modern hardware platforms as they are: platforms.