+ OpenAI's First Gadget Is A Screen-Free Speaker, And It Slipped To 2027
Reporting this week revealed that OpenAI's first hardware product, built with former Apple design chief Jony Ive, is a mobile speaker with no screen, meant to sit in your home and act like a companion. It answers questions, plays media, replies to messages, and controls smart home devices, all powered by ChatGPT. It has moving mechanical parts designed to make it feel alive and a camera that can take in its surroundings, like objects on a nearby table. The expected price is 200 to 300 dollars.
OpenAI bought Ive's hardware startup for roughly 6.5 billion dollars last year, and this speaker is the first product to come out of it. The full public release has now slipped to 2027 rather than this year. Part of the reason is a lawsuit: Apple sued OpenAI last week, claiming the company stole trade secrets tied to this exact device. OpenAI denies it.
What it all means: Consumer AI is trying to move off the screen and into a physical object that lives in your house and watches the room. The slip to 2027 and the lawsuit show how much harder and more contested that leap is than shipping another app. The company that defined the phone is now fighting the company that defined the chatbot over what sits on your counter next.
+ Google's AI Chief Called For A National Referee For Powerful AI
On July 14, Demis Hassabis, who runs Google's AI lab and shared a Nobel Prize for his work, published a plan calling for the United States to create a new watchdog with the power to review the world's most advanced AI models before they are released. He wants it modeled on the body that polices Wall Street brokers, funded by the industry but overseen by the government, and run by a majority-independent board of top scientists alongside company, government, and open-source voices.
Under the plan, AI labs would start by voluntarily sharing new frontier models with the body up to 30 days before release. Once the review process proved itself, passing it could become a requirement to sell a model in the US, and it would apply to every frontier model regardless of what country built it or whether it is open or closed. Hassabis wants it running before the end of the year and has spent months briefing the White House, rival lab leaders, and European officials.
What it all means: Two weeks ago the government quietly won early access to Anthropic's top model as the price of turning it back on. Now the head of Google's AI lab is proposing to make that kind of pre-release review a permanent, industry-wide institution, and to do it before year end. When the people building the technology start asking to be checked, the era of shipping first and explaining later is closing.
+ Big Tech Rivals Agreed On A Shared Standard For AI Assistants
This week Google and a long list of competitors announced a common standard that lets AI assistants from different companies find, verify, and use each other's tools and services. The backers include Microsoft, GitHub, NVIDIA, Salesforce, ServiceNow, Snowflake, Databricks, Cisco, and Hugging Face. Think of it as a shared directory and a set of manners: a way for one company's AI to look up what another company's AI can do and safely connect to it.
The move is widely read as the rest of the industry lining up to make sure no single company owns the wiring that connects all these AI tools together, with Anthropic and OpenAI the two names not at the table. Early versions are already live inside GitHub's and Hugging Face's tools.
What it all means: Rivals almost never agree on a standard unless they are afraid of the alternative. Agreeing on how AI assistants talk to each other is how this group keeps any one company from becoming the toll booth every other tool has to pay. For the businesses that will eventually buy these assistants, this is a quiet fight over which one can reach the software you already use.
