For years, texting between iPhones and Android devices was essentially a postcard — anyone with access to the right infrastructure could read it. That changes now.
Apple quietly dropped iOS 26.5 this week, and buried inside the update is something privacy advocates have wanted for a long time: end-to-end encrypted RCS messaging between iPhone and Android users. When it works, neither Apple nor Google can read what you're sending. That's a genuinely big deal, and it arrived with almost no fanfare.
The feature is still in beta, so it won't show up for everyone immediately. Apple says encryption will roll out automatically to new and existing RCS conversations over time, and Android users will need to be running the latest version of Google Messages for it to kick in. When it does, you'll see a lock icon and a small "Encrypted" label at the top of the conversation — a subtle but meaningful signal that your cross-platform chat actually has some teeth.
To understand why this matters, a little context helps. RCS was supposed to be the upgrade from SMS — better media sharing, read receipts, typing indicators. Apple adopted it last year after years of resistance, but that first version came without encryption for cross-platform chats. So while two iPhone users talking over iMessage had solid end-to-end protection, the moment an Android user entered the picture, that protection evaporated. This update closes that gap.
The encryption standard being used here is based on the GSMA's new technical specification for RCS end-to-end encryption, which Google helped develop and has had running on its end for a while. Apple's adoption is the missing piece that finally makes it a real cross-platform standard rather than a one-sided promise.
Now, the fine print. Carrier support is required, which means your mileage may genuinely vary depending on who your phone plan is with. Apple hasn't published a list of supported carriers, so there's a real chance some users flip to the update and see no difference at all — at least for now.
The rest of iOS 26.5 is a mixed bag. New wallpapers showed up, which nobody will complain about. But Apple also started serving ads inside Maps, which is the kind of move that raises eyebrows. The company says location data used for those ads isn't linked to your Apple account, but the arrival of ads in a navigation app is the sort of thing that tends to get worse over time before it gets better.
Apple also pushed matching updates across its entire product line — iPad, Mac, Apple Watch, Apple TV, and Vision Pro all received their respective 26.5 versions on the same day. The encrypted RCS feature, for now, is the one worth paying attention to.
AI
Mira Murati Reveals Thinking Machines Is Building Real-Time Interaction AI
Most AI models are, at their core, waiting machines. They sit idle until you finish your thought, then they respond, then they go quiet again. Mira Murati thinks that's the fundamental problem holding AI collaboration back — and her new company has a specific plan to fix it.
Thinking Machines, the AI lab Murati founded in February 2025 after her high-profile departure from OpenAI, announced this week that it's developing what it calls "interaction models." The concept is straightforward in theory and apparently very hard in practice: AI that continuously processes audio, video, and text in real time, responds as things unfold, and doesn't freeze up while it's generating a reply.
The current model for human-AI interaction has a structural bottleneck. You type or speak, the model waits, you finish, the model thinks, the model responds, and only then does it become aware of anything new. Thinking Machines compares this to trying to resolve an urgent disagreement over email instead of in person. You lose tone, timing, and the kind of back-and-forth that actually moves things forward.
What Thinking Machines is describing would work more like a conversation with a person in the room. The AI would notice if you trail off, catch changes in your tone, see what you're looking at, and respond to things as they happen rather than after a formal turn is completed. The company showed a few early demos: an AI that flags mentions of animals as a story is read aloud, a real-time speech translation tool, and a posture monitor that tells you when you're slouching. That last one is either the most useful or most irritating thing in tech depending on your personality.
The ambition here is bigger than any single demo suggests. Murati spent years at OpenAI overseeing the development of models that tens of millions of people use every day. Her argument with Thinking Machines seems to be that even the most advanced models today are operating through an unnecessarily narrow channel — and that the interface is as much of a limitation as the underlying intelligence.
That said, you can't actually try any of this yet. Thinking Machines says a limited research preview is coming in the next few months, with a wider release planned for later this year. In a space where announcements often race ahead of working products, that timeline deserves a healthy dose of skepticism.
The company has also had a rocky few months internally. Several early team members have left, with some heading to Meta and others returning to OpenAI — which is an uncomfortable detail for a lab that hasn't shipped a public product yet. Building a team that stays is its own challenge, and Thinking Machines will need to do that while also delivering on a genuinely ambitious technical vision.
Source: The Verge
Thinking Machines, the AI lab Murati founded in February 2025 after her high-profile departure from OpenAI, announced this week that it's developing what it calls "interaction models." The concept is straightforward in theory and apparently very hard in practice: AI that continuously processes audio, video, and text in real time, responds as things unfold, and doesn't freeze up while it's generating a reply.
The current model for human-AI interaction has a structural bottleneck. You type or speak, the model waits, you finish, the model thinks, the model responds, and only then does it become aware of anything new. Thinking Machines compares this to trying to resolve an urgent disagreement over email instead of in person. You lose tone, timing, and the kind of back-and-forth that actually moves things forward.
What Thinking Machines is describing would work more like a conversation with a person in the room. The AI would notice if you trail off, catch changes in your tone, see what you're looking at, and respond to things as they happen rather than after a formal turn is completed. The company showed a few early demos: an AI that flags mentions of animals as a story is read aloud, a real-time speech translation tool, and a posture monitor that tells you when you're slouching. That last one is either the most useful or most irritating thing in tech depending on your personality.
The ambition here is bigger than any single demo suggests. Murati spent years at OpenAI overseeing the development of models that tens of millions of people use every day. Her argument with Thinking Machines seems to be that even the most advanced models today are operating through an unnecessarily narrow channel — and that the interface is as much of a limitation as the underlying intelligence.
That said, you can't actually try any of this yet. Thinking Machines says a limited research preview is coming in the next few months, with a wider release planned for later this year. In a space where announcements often race ahead of working products, that timeline deserves a healthy dose of skepticism.
The company has also had a rocky few months internally. Several early team members have left, with some heading to Meta and others returning to OpenAI — which is an uncomfortable detail for a lab that hasn't shipped a public product yet. Building a team that stays is its own challenge, and Thinking Machines will need to do that while also delivering on a genuinely ambitious technical vision.
Enjoyed this?
Get stories like this delivered every Tuesday — free.