Recently I published a post titled ‘All My Apple AI Fears Just Came True’ where I expressed concerns that Apple was just trying to be another me-too AI company and how it was so un-Apple of them.
In the article I said Apple was being hypocritical because their servers have all the same problems they complained about other servers. And I got a comment that this was not the case and Apple was actually taking privacy very seriously. It linked to Apple’s white paper on the topic.
I’m not that convinced by the white paper. Because it’s like, “Don’t worry, the code is signed.” Like nothing bad has ever come out of signed code before.
Also signing an app is easy because it’s fairly small and has to be downloaded before use. So we can just hash the entire app and use public key cryptography to check this matches the known good signature. How do you sign an entire server with an LLM running on it? 🤷
Most of the techniques Apple talks about to prevent an attacker are more obfuscation than true security. Like this:
We supplement the built-in protections of Apple silicon with a hardened supply chain for PCC hardware, so that performing a hardware attack at scale would be both prohibitively expensive and likely to be discovered.
