Apple’s splashy announcement at its Worldwide Developers Conference this week, revealing the addition of artificial intelligence (AI) to its products and a partnership with ChatGPT-maker OpenAI, has sparked numerous questions about the functionality and privacy implications of these AI offerings.
The confusion is understandable. Apple is simultaneously launching a proprietary suite of AI models while integrating ChatGPT into its devices and software. It’s natural to wonder where one ends and the other begins and — perhaps more pressingly — what both companies will do with the personal information they receive from users. The stakes are particularly high for Apple, a company that has made security and privacy a hallmark of its brand.
The difference between Apple Intelligence and ChatGPT
Apple Intelligence — the collective brand name for all of Apple’s own AI tools — is intended to be more of a personal assistant than anything else, with an emphasis on “personal.” It takes in specific information about your relationships and contacts, messages and emails you’ve sent, events you’ve been to, meetings on your calendar and other highly individualized bits of data about your life. It uses that data to, Apple hopes, make your life a little easier — helping you dig up a photo you took from that concert years ago, finding the right attachment to put on an email, or ranking your mobile notifications by priority and urgency.
But while Apple Intelligence might know that you went on a hiking trip last year, it will lack what company executives called “world knowledge” — more general information about history, current events and other things that are less directly linked to you. That’s where ChatGPT comes in. Users will be able to have Siri forward questions and prompts to ChatGPT — on an opt-in basis — or have ChatGPT help you write documents within Apple apps. Apple plans to integrate with other third-party AI models eventually, too. The integration essentially removes a step to accessing ChatGPT and gives Apple users a more seamless onramp to that platform.
What does this mean for your data?
Since Apple Intelligence and ChatGPT will be used for largely different purposes, the amount and type of information users send to each AI may be different, too.
Apple Intelligence will have access to a wide range of your personal data, from your written communications to photos and videos you’ve taken to a record of your calendar events. There doesn’t seem to be a way to prevent Apple Intelligence from accessing this information, short of not using its features; an Apple spokesperson didn’t immediately respond to questions on that topic.
ChatGPT won’t necessarily or automatically have access to your highly personal details, although you might choose to share some of this data and more with OpenAI if you decide to use ChatGPT through Apple. In Monday’s demo, Apple showed Siri asking the user for permission to send a prompt to ChatGPT before doing so.
As part of its agreement with Apple, OpenAI made an important concession: OpenAI agreed not to store any prompts from Apple users or to collect their IP addresses — though all bets are off if you consciously decide to log in and connect an existing ChatGPT account. Some users might choose to do that to take advantage of their ChatGPT history or the benefits associated with ChatGPT’s paid account plans.
Should you trust Apple with your data?
Now that we’ve established what OpenAI will and won’t do with your data, what about Apple?
While Apple users will have to send their personal information and AI queries to OpenAI if they want to use ChatGPT, Apple has said that most of the time Apple Intelligence won’t be sending user data anywhere. As much as possible, Apple will try to process AI prompts directly on your device using smaller AI models.
This is similar to how Apple already processes FaceID and other sensitive data — the idea being that processing data right on the device limits risky exposure. Your data can’t be intercepted or hacked from a central server if it never actually goes anywhere.
In the event your AI task needs more processing power, Apple Intelligence will send your query and data to a cloud computing platform controlled by Apple, where a more capable AI model will fulfill the request.
This is where Apple claims it has achieved a major privacy breakthrough. The company’s announcement received relatively little airtime during its jam-packed Monday keynote but the company is plainly proud of the advancement it’s clearly been extensively planning.
Apple said Monday it has developed a new way to do cloud computing that means Apple can run computations on sensitive data while ensuring that nobody, not even the company itself, can tell what data is being processed or what computation is being done. Known as Private Cloud Compute, Apple’s new architecture borrows certain hardware and security concepts from the iPhone, including the secure enclave that already protects sensitive user data on Apple mobile devices.
With Private Cloud Compute, “your data is never stored or made accessible to Apple,” said Craig Federighi, Apple’s SVP of software engineering, during Monday’s keynote. After fulfilling a user’s AI request, Private Cloud Compute scrubs itself of any user data involved in the process, Apple said.
Apple claims Private Cloud Compute is “only possible” because of the tight control it has over its entire technology ecosystem — from the specialized, proprietary computer chips to the software tying everything together.
If it’s true that Apple can’t see the personal data that its large AI models are crunching — a claim Apple invited researchers to test for themselves because the system’s design is meant to be scrutinized — then that sets Apple’s implementation apart from that of other companies. For example, when you use ChatGPT, OpenAI discloses that it uses your data to further train its AI models. With Private Cloud Compute, you theoretically won’t have to take Apple’s word that it doesn’t use your data for AI training.
What about Apple’s training data?
Apple’s AI models didn’t spring up out of nowhere. They had to be trained, too, just like models offered by other companies. And that raises questions about whose data Apple used, and how. In a technical document released this week, Apple said its models are trained “on licensed data, including data selected to enhance specific features.”
“We never use our users’ private personal data or user interactions when training our foundation models,” the company added, “and we apply filters to remove personally identifiable information like social security and credit card numbers that are publicly available on the Internet.”
Apple did, however, admit to scraping the public internet for data that then went into training its proprietary models, making it somewhat similar to other AI companies, some of whom have faced copyright lawsuits and sparked a debate over whether AI startups have unfairly profited off the work of humans.
Apple hasn’t said what web-based information it has ingested. It did say that publishers can add code to their sites to prevent Apple’s web crawler from collecting their data. But that places the burden squarely on publishers to safeguard their own intellectual property, not the company.
