| ▲ | kmeisthax 15 minutes ago | |
LLMs can hack into a device but they can't keep a secret. For garbage IoT-tier crap, Claude can absolutely get into your device that you own. But for devices where the vendor has the economic desire to actually maintain effective control over your device you need to be able to stockpile bugs to retain access even after a patch. This is the same sort of "theoretically impossible but plausibly doable" logic that gives us the CIA's NOBUS policy - i.e. "security vulnerabilities only we know about because it's a pain in the ass to find them". What an LLM does to this is make NOBUS actually impossible, because anyone can spend $200/mo on a Claude pro plan and get actually novel vulnerabilities out of it. Great if you're worried about the CIA, terrible if you want to steal general purpose computing from a device vendor. What is actually happening is that bugs are getting leaked by prompt kiddies and patched before viable exploit chains can be made for them. I mean, think about it: Google is one of the biggest users of AI. You don't think they aren't going to be burning tokens - which they have access to at cost - on finding vulns before you can exploit them? | ||