Here’s what’s on our plate today:

• 🧪 Why AI is still searching for a body.

• 📰 ChatGPT's dangerous medical advice, IBM says AI isn't killing the mainframe, Samsung and Google's smart glasses.

• 💡 Roko's Pro Tip: know which parts of your product break if the platform changes the locks.

• 🗳️ Poll: whose vision for AI's body wins?

Let’s dive in. No floaties needed…

The Laboratory

TL;DR

• The body problem: AI companies promise a new way to deal with machines, yet it still reaches people through a phone designed before it existed.

• Lawsuit as tell: Apple's July 10, 2026 trade secrets suit against OpenAI, and the leak of a screenless home speaker days later, show how much rides on hardware nobody can buy.

• Three answers: Apple bets on the phone you own, NVIDIA on rebuilding the laptop from the silicon up, OpenAI on a new object entirely.

• Why it's hard: Humane raised over $230M and was sold for parts, the models still run in the cloud, and hardware talent sits mostly at Apple.

• The stakes: whoever owns the object people meet AI through owns the relationship, the data, and the gate every rival must pass.

Why AI is still searching for a body

For the past two years, nearly every argument about artificial intelligence has been an argument about the brain. The industry debated how large the models could grow, whether the circular financing behind them, in which the same companies fund and buy from one another, could hold, and how far regulators should reach into their training. What almost nobody debated was the body. AI still reaches people the way software always has: as an app on a phone or a tab in a browser, living inside hardware designed a decade before the technology existed.

The mismatch is strange because the companies building AI have never sold it as another app. They have sold it as a new way for humans to deal with machines altogether, one where you talk and the technology understands, watches, and acts, a promise far too large to keep through a keyboard and a grid of icons. Even so, the question of what physical object would actually carry AI into daily life went largely unexamined. Part of the reason was that the few companies willing to answer it discovered how difficult the problem really was.

The hardware problem

Humane, a startup founded by former Apple designers, raised more than $230M to build a wearable pin that would replace the phone. Within a year of shipping, its returns were outpacing its sales. HP bought its remains for $116M, and every pin was permanently turned off. The lesson the industry drew was that AI hardware was a graveyard, and the body question largely went quiet. This month, however, it returned in an unlikely form: a lawsuit and a leak.

On July 10, 2026, Apple sued OpenAI in a California federal court, accusing it of stealing trade secrets to build consumer hardware of its own. The complaint alleges a coordinated effort to pull designs, supplier lists, and even physical components out of current and former Apple employees, and it notes that more than 400 of them now work at OpenAI, many recruited into the hardware group OpenAI built by acquiring io Products, the startup co-founded by former Apple design chief Jony Ive, for $6.5B. Four days after the filing, reporting revealed what all this secrecy is protecting: OpenAI's first device is said to be a portable, screenless speaker with a camera and sensors, a machine pitched internally as a humanlike companion for the home, expected to be unveiled this year and shipped in 2027. None of it has been shown publicly, and the lawsuit itself could push the timeline.

Yet OpenAI is not the only company attempting to decide where AI will live. Just a few weeks earlier, a quieter announcement hinted at a very different answer. NVIDIA unveiled a chip that puts its AI processors inside ordinary Windows laptops, built with MediaTek. The chip is reportedly headed for machines from Dell and Lenovo, and NVIDIA paired it with a Microsoft partnership to run AI agents, software that carries out tasks on a user's behalf, directly on the PC rather than in a distant data center.

Neither announcement was really about hardware specifications. Both answered the same question: if AI becomes the primary interface between people and computers, where should it live? Three visions of where AI will live are now in open competition. Apple's answer is the device you already own, with intelligence folded into the phone. NVIDIA's answer is also the device you already own, rebuilt from the silicon up, so the laptop itself becomes the AI machine. OpenAI's answer is a new object entirely, one that has no screen because it is not meant to be looked at. None of these objects yet exists in a form a person can buy, which is what makes the intensity of the fight over them, waged through poaching raids and a 41-page complaint, so striking.

The fact that three of the world’s largest technology companies disagree so fundamentally about AI’s future home points to a deeper contradiction beneath the entire debate. The AI industry's central promise is that the technology will change how humans interact with machines, yet every attempt to give that change a physical form has failed, while the form it is supposedly escaping keeps winning.

The body we already have

The smartphone is the most refined consumer object ever made; it is already in everyone's pocket, and it runs every AI assistant perfectly well. A reasonable observer could conclude that the phone is simply AI's body, and that the search for a new one is a solution in search of a problem.

Humane’s collapse is the strongest evidence for that view, but it is difficult to square with the behavior of the companies building frontier AI. OpenAI, after all, is behaving as though the smartphone is precisely the problem. The counterargument is what OpenAI's behavior implies: a company that believed the phone was enough would not spend $6.5B on designers, absorb hundreds of Apple engineers, and accept a bruising lawsuit to build a speaker. Living inside someone else's hardware means living by their rules, paying their fees, and reaching users only through the narrow channel of an app. For a company that wants AI to be the platform rather than a feature, the borrowed body is a cage, however comfortable it may be.

Escaping the cage

OpenAI understands that escaping the cage is punishingly hard, as evidenced by the graveyard of failed hardware. A new device must beat a supercomputer that people already carry, love, and have organized their lives around, and it must do so on day one, without the apps, the accessories, or the muscle memory.

However, the physics are unforgiving: powerful models still mostly run in the cloud, so a standalone gadget inherits every network delay and dies the moment its servers do, as Humane's customers learned when their pins were bricked by an email. The supply chain runs through the same few Asian manufacturers Apple has spent two decades binding to itself, which is partly why the lawsuit dwells on supplier lists. And the talent that knows how to ship 100M polished objects works overwhelmingly at one company in Cupertino, which is why the fight opened over people rather than products.

NVIDIA’s path is the revealing exception, because it sidesteps every one of these walls by upgrading the bodies people already own instead of asking them to adopt a new one, a quieter bet that for now carries the least risk. Those obstacles help explain why the search for AI’s body remains unresolved. They also explain why NVIDIA’s answer is so different from OpenAI’s.

The prize

The ferocity makes sense once the prize is named: whoever owns the object through which people meet AI owns the relationship itself, the data flowing through it, the payments riding on it, and the gate every other company must pass through. Apple has run that tollbooth for nearly two decades, and the lawsuit reads less like a complaint about files than a defense of the booth. The deeper question the fight raises is one of direction. Either AI keeps adapting itself to hardware built for a pre-AI world, or hardware starts reorganizing itself around what AI can do, and each path produces a different decade of computing.

For all the lawyers and the leaks, the story returns to the box. The brain has never been more capable, and it still lives where it started, behind glass, waiting to be tapped. The body meant to free it is still a rumor; its shape is known only from reporting and the allegations in Apple's filing, defended by lawyers before a single customer has touched it.

It may ship in 2027 and change what a computer is. It could join the pin in the drawer of things that tried. The one certainty the past fortnight offers is that the largest technology companies no longer consider the question small. Whether the first body AI is a new machine or simply a lawsuit deciding who is allowed to build one remains open.

Roko Pro Tip

Monday Poll

Bite-sized Brains

• ChatGPT's dangerous medical advice: A new lawsuit alleges ChatGPT gave harmful medical advice, sharpening the fight over who's liable when AI gets high-stakes answers wrong.

• IBM says AI isn't killing the mainframe: After a shocking quarter, IBM is insisting AI isn't replacing the mainframe, pushing back on the idea that legacy infrastructure is on its way out.

• Samsung and Google's smart glasses: Samsung and Google are teaming with Gentle Monster and Warby Parker on smart glasses, betting that eyewear brands are the key to making the category mainstream.

Meme Of The Day

The Toolkit

• Assembly AI: Speech-to-text API that handles transcription, speaker detection, and audio intelligence for production apps. 

• Chroma: Open-source vector database built for AI apps, fast to set up and easy to scale for RAG and embeddings. 

• Continue: Open-source AI code assistant that plugs into VS Code and JetBrains with full control over models and context.

Rate This Edition