Apple’s AI and privacy: why Elon Musk is sounding the alarm

Credit: Apple

Only a few hours after Apple took the wraps off its brand new Apple intelligence-infused software, billionaire Elon Musk has already raised privacy concerns.

In a series of scathing messages on his X messaging platform, Musk’s claimed that Apple’s partnership with Open AI’s ChatGPT leaves users’ privacy vulnerable.

Musk also noted that Apple should have developed its own version of ChatGPT and not relied on partnering with OpenAI.

Musk even went as far as threatening to ban iPhones from his company’s offices: “If Apple integrates OpenAI at the OS level, then Apple devices will be banned at my companies. That is an unacceptable security violation.”

On the other hand, Apple’s new AI capabilities have been designed from the ground up to protect privacy, so do Musk’s claims hold any weight?

Is using AI a privacy risk?

First, a bit of background. For AI to be truly useful on a device like your iPhone, it needs to be able to access your personal information. Apple Intelligence, which is the name for the company’s new AI framework, continuously scans your phone’s data including emails and text messages, calendars, contacts and other data so it can understand what’s important to you. 

It stores this in a special database called a ‘semantic index’, and this is essential to providing context to your requests. As you would expect, the semantic index file contains personal information that needs to be kept safe from prying eyes.

Ultimately, any AI will either need access to your data or produce information about you based on your requests, so careful consideration needs to be given about how this is protected.

How Apple Intelligence protects your privacy

For Apple Intelligence to do something for you, it starts by determining how and where your requests should be handled. The current options are: on-device, on Apple’s servers or with ChatGPT, but more on this later.

The most fundamental way to protect your privacy is to do the ‘thinking’ or processing ‘on-device’, meaning your iPhone (iPad or Mac). 

This is much safer than sending your data across the internet to other computers, where it could become vulnerable to data breeches, questionably privacy policies and other risks.

Thankfully, Apple has designed the hardware and software on its latest iPhones, Macs and iPads to be capable of handling most AI-related computational tasks ‘on-device’.

Apple’s Private Cloud Compute

For more complicated work that needs more processing grunt or space to store large “learning models”, some requests still need to be routed to Apple’s powerful computers located in the cloud.

This is where Apple’s new Private Cloud Compute system take centre stage. Private Cloud Compute is a cluster of powerful computers made using Apple’s chips (silicon), specially designed to manage complex computational tasks sent from many devices at once.

And since Private Cloud Compute uses Apple’s exclusive silicon, there are unique protections in place not found on other cloud-based AI systems. This including the use of a special chip called the ‘Secure Enclave’’, which also happens to be found your iPhone, iPad or Mac. This makes it possible authenticate very secure encrypted connections between your device and the server it’s sharing information with.

Image of ChatGPT being used in Apple interface
Some may prefer to have ChatGPT directly integrated into Apple Intelligence instead of accessing it from a website. Credit: Apple

Apple also minimises the information that’s sent to the cloud, so, rather than sharing the contents of your emails, for example, it would only send details relevant to the task at hand.

And since the task’s results and data are not stored on the server, Apple can’t access any information, and it can’t fall into the wrong hands.

And lastly, Apple has included a mechanism for security experts to easily examine and audit how well your privacy is protected by Private Cloud Compute.

The ChatGPT risk

And then there’s the third and most contentious option, and the one that Elon Musk has the biggest issue with. Apple, presumably still not 100% happy with its own AI’s capabilities, has partnered with OpenAI to use its ChatGPT4o manage some types of requests.

These are tasks that ChatGPT is very good at, such as document handling, writing prose to a specific style, or some generative image tasks.

Some may also prefer a direct integration of ChatGPT into Apple Intelligence instead of having to log into ChatGPT separately from a web browser in order to access the service.

When it comes to privacy when using ChatGPT, Apple says that there are protections in place. This includes hiding your identity by concealing your IP address and only sending request data to ChatGPT.

Also, Apple will prompt you when it decides that ChatGPT is the best option for handling your task, so you can decide not to use it. 

I’ve also learned that if you are not logged in a ChatGPT account, your requests are completely anonymous. If you have a ChatGPT account however, it will store the history of your conversation to your account, as it normally would when accessing ChatGPT directly. This would provide the benefit of context should you want to reference your request in future queries.

However, the big concern is OpenAI’s dubiously worded privacy policy, which casts doubts about how well it protects information that it might learn about you. 

ChatGPT’s data-use policies apply for users who choose to connect their account

Apple website

Also, the company has recently become vulnerable to exploits that reveal private user’s data, and lastly, Musk’s rocky relationship with OpenAI over it becoming a ‘for profit’ entity certainly doesn’t put it in his good books.

Beyond this, there’s also the question that as the use of ChatGPT is banned by many companies, including Apple, why is it OK for it to be integrated across Apple devices?

In the long term, ChatGPT may be replaced as Apple Intelligence becomes more capable, which might make Musk a little more accepting of Apple’s AI system.

It’s also reassuring that some new privacy tools will be available in a few months with the latest versions of iOS, MacOS and iPadOS, including lockable apps, hidden folders, a new Passwords Manager app, a revamped Privacy and Security settings summary, better control of contacts and more.

Valens Quinn attended WWDC 2024 as a guest of Apple Australia