The era of artificial intelligence everything is here, and with it, come everyday surprises into exactly where the next AI tools might pop up. There are major corporations pushing customer support functions onto AI chatbots, Big Tech platforms offering AI image generation for social media posts, and even Google has defaulted to include AI-powered overviews into everyday searches. The next gold rush, it seems, is in AI, and for a group of technical and legal researchers at New York University and Cornell University, that could be a major problem. But to understand their concerns, there’s some explanation needed first, and it starts with Apple’s own plans for AI. Last October, Apple unveiled a service it is calling Apple Intelligence (“AI,” get it?), which provides the latest iPhones, iPads, and Mac computers with AI-powered writing tools, image generators, proof-reading, and more. One notable feature in Apple Intelligence is Apple’s “notification summaries.” With Apple Intelligence, users can receive summarized versions of a day’s worth of notifications from their apps. That could be useful for an onslaught of breaking news notifications, or for an old college group thread that won’t shut up. The summaries themselves are hit-or-miss with users—one iPhone customer learned of his own breakup from an Apple Intelligence summary that said: “No longer in a relationship; wants belongings from the apartment.” What’s more interesting about the summaries, though, is how they interact with Apple’s messaging and text app, Messages. Messages is what is called an “end-to-end encrypted” messaging app. That means that only a message’s sender and its recipient can read the message itself. Even Apple, which moves the message along from one iPhone to another, cannot read the message. But if Apple cannot read the messages sent on its own Messages app, then how is Apple Intelligence able to summarize them for users? That’s one of the questions that Mallory Knodel and her team at New York University and Cornell University tried to answer with a new paper on the compatibility between AI tools and end-to-end encrypted messaging apps . Make no mistake, this research isn’t into whether AI is “breaking” encryption by doing impressive computations at never-before-observed speeds. Instead, it’s about whether or not the promise of end-to-end encryption—of confidentiality—can be upheld when the messages sent through that promise can be analyzed by separate AI tools. And while the question may sound abstract, it’s far from being so. Already, AI bots can enter digital Zoom meetings to take notes. What happens if Zoom permits those same AI chatbots to enter meetings that users have chosen to be end-to-end encrypted? Is the chatbot another party to that conversation, and if so, what is the impact? Today, on the Lock and Code podcast with host David Ruiz, we speak with lead author and encryption expert Mallory Knodel on whether AI assistants can be compatible with end-to-end encrypted messaging apps, what motivations could sway current privacy champions into chasing AI development instead, and why these two technologies cannot co-exist in certain implementations. “An encrypted messaging app, at its essence is encryption, and you can’t trade that away—the privacy or the confidentiality guarantees—for something else like AI if it’s fundamentally incompatible with those features.” Tune in today. You can also find us on Apple Podcasts ,
