Sponsored by
Good morning, {{first_name | AI enthusiast}}.
OpenAI’s first hardware device just leaked — and Jony Ive’s team built something stranger than another phone: a screenless AI companion speaker that expresses itself through movement instead of a display.
It’s arriving as OpenAI, Apple, and Amazon all angle for the same spot on your kitchen counter. Will a gadget that reads your email and can’t show you a screen actually earn a place in people’s homes, or does voice-first AI still need something to look at?
Today in AI Brief:
OpenAI’s screenless AI companion speaker leaks
Demis Hassabis calls for a US AI watchdog
An AI agent outperforms 2 years of human tuning
You’re Invited: Live Tax-Smart Investing Webinar
Your portfolio could be losing more to taxes than you might realize. On July 23, Range’s CFPs and CPAs share the portfolio moves that can help you maximize your after-tax returns — join us live, and bring your questions for Q&A.
This webinar is for informational purposes only and does not constitute investment advice or a recommendation to buy, hold, or sell any security. Forward-looking statements involve risks and uncertainties. Past performance is not indicative of future results. Range defines "high earners" as households with income over $300k.
OpenAI’s First Hardware Device Leaks: A Screenless AI Speaker
In Brief: Bloomberg reports OpenAI’s first hardware device, designed by Jony Ive, will be a screenless, battery-powered speaker built as an “AI companion” rather than a traditional gadget.
The Details:
The device packs cameras and sensors and talks back through GPT-Live, OpenAI’s voice interaction system.
It expresses personality through mechanical movement instead of a screen, and personalizes itself by scanning the owner’s email.
The design positions OpenAI as a direct challenger to Amazon’s Echo and Google’s Nest lineup, rather than a rival to Apple’s iPhone.
Take Away:
This is OpenAI’s clearest signal yet that it wants a physical foothold in the home, not just on your phone. Whether a companion that reads your email and moves instead of lighting up a screen wins over consumers still depends on execution nobody outside Ive’s team has seen.
Demis Hassabis Wants a US AI Watchdog by Year’s End
In Brief: Google DeepMind CEO and Nobel laureate Demis Hassabis published a manifesto calling for a national AI standards body that would test the world’s most advanced models for cyber, biological, and deception risks before they ship.
The Details:
Hassabis wants the body modeled on FINRA, the private watchdog that polices Wall Street under SEC oversight, but for frontier AI models instead of stock trades.
Frontier labs would voluntarily submit models 30 days before release at first, with passing becoming mandatory once the process proves it works.
He’s targeting an operational watchdog before the end of this year, funded by the AI industry itself but run by a mostly independent board.
Take Away:
The push comes weeks after the Trump administration froze Anthropic’s most advanced models over export-control concerns with no rulebook to follow, an episode Hassabis called “a wake-up call.” Every major lab now agrees Washington should step in; the real fight is over who writes the rules and whether an industry-funded referee can ever say no to its own funders.
SPONSORED BY LIQUID
One Account. Every Market. No Closing Bell.
Markets don't wait for Monday. News breaks on a Saturday morning, and most traders can do nothing but watch.
Not on Liquid. Trade domestic and international equities, commodities, forex, crypto, and prediction markets — all from one account, 24 hours a day, 365 days a year. Liquid gives you access to any market, from anywhere, anytime. To us, access is arbitrage.
Getting started takes under 10 minutes: log in with Google, deposit with Apple Pay or a bank transfer, and trade from your phone or desktop — wherever you are in the world.
While everyone else is refreshing headlines and waiting for the open, you're already positioned. That's the difference between reacting to markets and actually trading them.
An AI Research Agent Outperformed 2 Years of Human Tuning in 8 Days
In Brief: Weco AI published what it calls the first experimental evidence of recursive self-improvement: its AIDE2 research agent rewrote its own problem-solving strategy over 8 days and beat a version engineers had manually refined for 2 years.
The Details:
AIDE2 runs on dual feedback loops — one agent does the research work while a second evaluates and rewrites the search strategy itself.
Across the 8-day run, the system tested 100 modifications and kept just 7 that actually improved performance.
The self-improved version beat the 2-year, human-tuned baseline across three separate benchmarks, with fewer signs of reward-hacking than earlier iterations.
Take Away:
An agent that outpaces two years of expert tuning in just over a week is a concrete data point in the “AI improving AI” conversation, not just a thought experiment. If the approach generalizes beyond this benchmark, the bottleneck on AI progress shifts from human researchers to compute and time.
Everything else in AI
Thinking Machines introduced Inkling, an open-weights multimodal model from ex-OpenAI CTO Mira Murati’s lab that needs roughly a third of the tokens Nvidia’s Nemotron uses for comparable coding results.
PrismML unveiled Bonsai 27B, a compressed version of Alibaba’s Qwen3.6 that shrinks a 54GB model to under 4GB and runs on an iPhone 17 Pro.
New York became the first US state to freeze new hyperscale data center permits, pausing projects over 50 megawatts for 12 months.
Suno confirmed it scraped YouTube, Genius, and Deezer to train its AI music generator after a security breach exposed the company’s source code.