Daybreak - OpenAI's next product is a companion. Its own research says that's the problem

Episode Date: July 22, 2026

OpenAI's own research found that its heaviest users were also its loneliest and most emotionally dependent. The top 10% kept triggering classifiers built to detect unhealthy attachment.The response was to build a screenless AI companion designed to live in your home, learn your habits, access your emails, and get harder to let go of over time.Meta, meanwhile, is selling AI glasses with Kylie Jenner's voice for an extra hundred dollars — and they sold out before the launch event ended.Both companies have documented the risks of AI dependency internally. Both are shipping anyway.Tune in.Daybreak is produced from the newsroom of The Ken, India’s first subscriber-only business news platform. Subscribe for more exclusive, deeply-reported, and analytical business stories.

Last week, a Bloomberg report revealed the details of Open AI's first-ever hardware product. Sometime next year, the company is going to sell you an AI companion that interacts with you in a human-like way and lives in your home. It's reportedly a portable speaker with no screen, but it does have some movable parts that don't do anything specific yet except to make you feel like the device is alive. It is also designed to have access to as much of you as possible and understand you more deeply over time and connect with you better. And Sam Altman has big plans for this device. It is currently being designed by Johnny Ive, who used to be the chief design officer at Apple.

And when Altman had acquired Ives AI hardware design startup called I.O. in mid-20205, Wall Street Journal reported that he had announced to his Open AI staff that he plans to ship 100 million AI companions with the aim that they would become a part of users' everyday life. But OpenAI isn't exactly the first company trying to sell users an intelligent companion recently. About a month ago, Meta launched its own line of AI glasses, this time integrated with its new AI model called Muse Spark. And one of the editions in this line called the Starfire Kylie was designed in collaboration with Kylie Jenner, who you might know as the Kardashian family entrepreneur and model.

This particular style comes with an additional perk for an extra $100. The AI voice on this edition alone sounds like Jenner herself. The irony is that just two years ago, OpenAI itself had published a system card for its popular GPT-40 model. A system card, by the way, is basically a research report that covers the safety concerns that come up during the testing phase. The report showed that users who use the voice mode of the AI model risked getting unhealthily attached to it. The researchers wrote, and I'm quoting here, that users might form social relationships with the AI reducing their need for human interaction, potentially benefiting lonely individuals, but possibly affecting healthy relationships.

So the industry is more than aware of the downside of introducing AI companions to the market. But Open AI is readying for an IPO. It has spent $6.5 billion just in acquiring I.O. And Altman has said that the acquisition will add $1 trillion in value to the company. Meta, on the other hand, has burned more than $80 billion on its reality labs. the same arm that has been working on the meta-ray bands and Oakleys with the glasses being the division's only popular product. It's already in discussions to increase production capacity by the end of this year.

Which means both companies are betting that attachment will save their bottom line, despite the documented risk. Welcome to Daybreak, a business podcast from the Ken. I'm your host, Trejewergies, and every day of the week, my co-host, Nika Sharmanah will bring you one new story that is worth understanding and worth your time. Today is Wednesday, the 22nd of July. The use of the word companion feels like a very deliberate turn from the assistant model that was

popular just a few years ago, like Alexa or Siri. Altman himself referred to the new devices as companions, not assistants. Zuckerberg also made headlines last year in a conversation with podcaster Duarkeesh Patel. Here is how the time put Zuckerberg's argument, and I'm quoting here. He pointed out that the average American has fewer than three friends and argued that the answer may not be in traditional policy solutions like reinvesting in community centers or expanding mental health support. I think people are going to want a system that knows them well and understands them in the way that their field algorithms do. He continued, envisioning AI friends

companions and therapists as the remedy for modern disconnection. Notice how Zuckerberg is also avoiding the word assistant. Friends, companions, therapists. This shift in vocabulary seems to be mirroring a larger change in what's happening in the way that people use AI. These words describe things that are there for you, not just the things that do things for you. For example, a research project by Mars General Brigham,

a nonprofit that runs two US academic medical centers reported that 87% of daily users use chat pots for personal reasons, which could include recommendations, advice, or emotional support. And this turned towards the use of companions makes more sense when you know what happened to the assistant era of devices. Alexa is probably the cautionary tale when it comes to AI assistance. A WSGA report two years ago, revealed that despite Alexa being in millions of households, Amazon had lost $25 billion in its devices business. When the company began selling its EcoSmart home devices

with the Alexa Voice Assistant at incredibly low prices, the idea was that it would eventually recover that money once customers started spending on shopping through the Voice Assistant. But that bet never paid off. Instead, what ended up happening was that customers used Echo mostly for the free apps for things like setting alarms and checking the weather. In fact, a former senior employee told WSJ that the company was worried that it had hired 10,000 people and basically all they had done was build a smart timer. Even before that, another employee had called Alexa a colossal failure of imagination and a wasted opportunity. In 2023, Amazon eventually ended up laying off several hundred people in its Alexa division

and focused resources on AI development instead. And up until recently, it seemed like Meta's Reality Labs was also on a similar path. But this time around, its latest product might actually be the light at the end of the tunnel. Its Reality Labs division of the company has been around since 2012. And its main aim has been to research and develop capabilities in the virtual reality, augmented reality and metaverse space. The company has seen a period of significant burn, reportedly about $80 billion over the last six years alone. And in January this year, Meta also laid off about 1,000 people from the labs, which is about 10% of the division's workforce. And during this time, the most successful product to come out of the division are the smart glasses.

The glasses have been so successful, in fact, that Meta and Ezellul Luxottica, the brand that owns Rayban and Oakley, are planning on increasing annual production capacity to 20 million units or more by the end of 2026. This decision, Business of Fashion Reports, is so that the company can capture growing demand and get ahead of its competitors. And even as it is reportedly deepening its relationship with Eselol Exotica, after the success of the Meta rebands and Oakleys in its latest launch, Meta designed its new range of glasses in-house.

And with variants like Starfire Kylie Edition, Meta is focusing on glasses that are described on the Meta Store website as exclusively packaged and made to feel personal from the moment that it arrives. Personal is the word. that is interesting here, because everything about the product is designed to be a fashion statement and an intimate one at that. Reportedly, the celebrity herself asked for a characteristic gemstone on the corner of a lens and even the charging case for the glass comes with a mirror for a quick check or a touch-up, of course.

Even the series of selfies Jenna posted on her Instagram to promote the product, show her in a bathrobe just posing in front of a mirror or applying her. makeup. The glass is deliberately shown as a part of private personal moments. And despite some serious online backlash, which criticised the model for lending support to an item that has become synonymous with surveillance tech, the glasses do seem to be flying off the shelf. An influencer who wrote for Yahoo creators wrote that by the time she got to a meta-creator event, the Starfire Kylie edition had sold out in every available color. She heard other words. She heard other attendees asked staff for the same edition by name only to get the same out-of-stock reply.

So clearly, even with the $100 premium, the campaign seems to be working. You know, Amazon had actually tried selling something similar back in 2019 when it sold Alexa users a Samuel L. Jackson voice pack for a dollar. It only lasted about three years before it was discontinued. But what's new with the Kylie classes isn't the idea of a celebrity voice. It's the sheer scale of her fame being sold, and the price tag, and also the fact that it's a piece of hardware that sits on your face and not your kitchen counter. On the other hand, OpenEI's device, even though it does sit on a kitchen counter, is also meant to get more intimate with time. Both Open AI and Meta's take on companion devices is solving the problem that Amazon faced. But they're also risking the creation of another one.

Stay tuned. months before that system card I mentioned earlier even came out, in May 2024 when the voice mode first launched, Sam Altman tweeted a single word, her. It was a reference to the 2013 film that starred Joaquin Phoenix about a lonely man who falls in love with his AI. Scarlett Johansson, who voiced the AI in the movie, wasn't at all flattered in real life.

She said Chad GPD's new voice sounded like her and that she had already turned down Altman's request to use it for real. And soon, Altman's joke turned out to be pointing at something his own company would confirm in writing three months later. That's when the system card I mentioned earlier came out. And it said that Chat JPT's voice mode users were more likely to form the kind of attachment that could actually hurt the relationships with real people. And now, in 2026, Open AI seems to be ignoring its own warning. Its new device takes the Alexa model, a screenless speaker that lives in your home, and deepens the relationship it can have with its owner.

The device doesn't just sound more human. It even acts more human. Reportedly, the voice model that goes in the speaker is capable of listening to a user and thinking at the same time, mimicking the cadence and rhythm of human conversation. Plus, like I mentioned earlier, it also has movable components that are specifically designed to make it feel more alive. And unlike an Alexa and like the meta glasses, it also comes with cameras and other sensors to understand its and the user's environment. It's also designed to get to know its user over time by having access to sensitive information like emails and other personal info, which means it will probably only get harder for you to let go off.

But why is attachment as a model suddenly so attractive to both companies right now? The answer may lie in this other study that OpenAI and MIT Labs ran together. They analyzed around 40 million chatypd conversations and ran a month-long trial with a thousand users. They found that the heaviest users were also the loneliest and more than. emotionally dependent ones. They also found that the top 10% of users kept setting off the classifiers that were built for the study to detect emotional attachment between the user and the chatbot, which means that the company's best customers and its most at-risk users are basically

the same people. And the warning isn't exactly a hypothetical. Unfortunately, there are examples of what adverse forms of this kind of dependency looks like in real life. I'd cover this in an episode on OpenE.I. Flotation with an adult mode in April. Character AI, a platform for AI chatbots, faced a wrongful death lawsuit, alleging that a chatbot had fostered harmful dependency in a teenager,

which had led him to taking his own life. A lawsuit filed by his mother claimed that he was in love with the bot and shared explicit chats with it. The company ended up blocking teens from accessing open-end chats and settle the lawsuit with the family. Now, OpenAI itself is also battling a wrongful death lawsuit from the family of a 16-year-old who passed away by suicide, alleging that Chad GPT fostered a psychological dependence in him. That's a case that's still active with OpenAI disputing any responsibility so far. Even Meta has a legal precedent that is working against it.

This March, a jury found that the company was negligent for designing Instagram to be addictive. to teenagers. Now, meta is appealing against this order, but the case has showed something familiar. Its own researchers had flagged the harm internally, but the company kept the design anyway. Now, with a voice-led device that lives in your home or on your face, this risk is only more likely to increase. But as far as a ticking dashboard of numbers is concerned, a dependent user and a loyal one who never unsubscribes looks pretty much the same. And even though Open AI ran the study itself, here is the uncomfortable part. Even if the company has a capacity to differentiate between a loyal user and a dependent one,

nothing actually requires it to. At least legally not yet. But with Anthropic filing for its own IPO just weeks before Open AI and Google alone planning to spend close to $200 billion on AI this year, neither company can really afford to slow down. and asks the hard questions about what exactly it is building and how it is going to affect the people it's going to sell to. Right now, both Meta and Open AI look like they've taken a known industry risk and turned it into a business model. Daybreak is produced from the newsroom of the Ken India's first subscriber-focused business news platform. What you're listening to is just a small sample of our subscriber-only offerings.

A full subscription offers daily long-form feature stories. newsletters and a whole bunch of premium podcasts. To subscribe, head to the ken.com and click on the red subscribe button on the top of the Ken website. Today's episode was hosted and produced by my colleague Rachel Vargis and edited by Rajiv Sien.