The Download

Illustration: Morning Brew Inc., Photo: Adobe Stock

TL;DR: Anthropic—not long ago seen as the underdog—has switched places with OpenAI on two fronts this week: revenue and safety. It goes to show how quickly the tides can turn in the AI race.

The revenue shake-up: For the first time ever, Anthropic brought in more revenue than its rival in the most recent quarter, according to the Wall Street Journal. The company posted $11.6 billion in sales in Q2, more than double what it made the previous quarter. OpenAI, meanwhile, racked up about $6.7 billion in the same period—up 18% from Q1, but overshadowed by its ballooning losses, which grew from $9.3 billion to $12.3 billion.

It’s a reversal of fortune from 2025, when OpenAI made about $13 billion in annual revenue compared to Anthropic’s roughly $10 billion. The ChatGPT maker recently said that it has over 1 billion active users now, but the majority enjoy the product for free, and growth has been slowing. Meanwhile, it’s been a boom time for Anthropic thanks to the ongoing popularity of Claude Code.

The safety switch-up: Revenue isn’t the only thing that’s flipped. Yesterday, hours before those numbers surfaced, OpenAI said it had paused some training on its unreleased Astra models and was slowing the pace of development overall. Sam Altman told reporter Alex Heath that “getting AI safety right is more important than any company’s momentum.”

It’s a line you’d expect from the head of Anthropic, long perceived as the more cautious, safety-conscious lab. Except Anthropic just said the opposite in a risk report, noting that there was no need to slow down its own training at this time. The announcements come after both companies have reported cybersecurity incidents involving AI agents gone rogue this summer.

How the tables turn: Anthropic, if you’ll recall, was founded by former OpenAI execs and researchers who left due to safety concerns—and since then, it’s been seen as a scrappy lab playing catch up to OpenAI. Now, Anthropic looks ascendant while its competitor is under visible strain, with reports that it’s facing more internal strife over safety.

Bottom line: In 2026, we’ve already seen Anthropic gain momentum and pull ahead in the AI vibes race. But these new developments put it squarely in pole position as the two companies hurtle toward mega-IPOs. —WK

      

Sponsored By Siemens Software

Agents aren’t the only AI that’s coming to the workforce. We recently teamed up with the robotics experts at Siemens to talk about humanoids.

As different industries look to augment their workforces, advances in technology have enabled humanoid robots to emerge as a viable way to do so.

So if you’re curious about what these robots are capable of, what industries are looking to implement them, and what tech is making it all possible, we’ve got those answers.

They’re all waiting for you in the article. Go ahead, give it a read, and get ready for robots in the workplace.

• ✈️ The largest electric plane used just $5 of electricity for a flight. The aircraft, which has a wingspan of 106 feet (about the size of a single-aisle plane), stayed in the air for 27 minutes and reached an altitude of 1,100 feet. According to its manufacturer, it generated more than 1 megawatt of power for the price of a latte. The plane is a demo of a hybrid-electric regional airliner built to carry up to 30 passengers that’s slated for flight testing in 2028 and entry into service in 2031—United Airlines, Air Canada, and JSX have all placed commitments.
• 💉 A new vaccine could prevent skin cancer from coming back. Moderna and its partner Merck said today that a personalized mRNA shot kept melanoma from coming back or spreading, based on results from a large, late-stage clinical trial. Here’s how it works: After a patient’s tumor is removed, scientists scan its genetic profile for mutations most likely to trip the immune system. Then, they encode those mutations into mRNA and manufacture a custom shot in a process that takes about six weeks. In the trial, more than 1,000 high-risk patients, whose detectable cancer had been removed through surgery, received the vaccine along with immunotherapy. —LC

The Zeitbyte

Cussing-edge technology

Morning Brew Inc.

AI labs are always boasting about new model improvements—but the latest version of ChatGPT is displaying an unbenchmarked skill: swearing like a sailor. Users started noticing its penchant for filthy language in the last few weeks, though they’re not complaining—many appear delighted to see the AI chatbot loosen up. (One person told Business Insider that, as an Australian, it made conversation feel more “natural.”)

Whether this is a sign of superintelligence, or the opposite, is anyone’s guess. The AI can sometimes sound like a middle schooler who just learned a naughty word, dropping a sudden f-bomb while explaining how much green onion to use in a recipe, per BI. OpenAI declined to comment for the outlet’s story, but when Tech Brew asked the company’s GPT-5.6 Luna model about reports of increased cursing, it told us that some users prefer a “more casual, unfiltered tone, and models can mirror that style when the conversation invites it.” Notably, it opted not to use any fudging profanity in its response. —WK

Chaos Brewing Meter: ☕☕☕/5

Sponsored By Currex

Your shoes called. They need backup. Europe’s best-kept insole secret is now America’s most-bought one. CURREX insoles are engineered by German sports scientists to fix how your foot actually moves—not just cushion it. Built for running, soccer, football, hoops, and whatever fall throws at you. Slip them in. Feel the difference before your first sprint.

• ICE banned its employees from wearing Meta smart glasses while working due to privacy concerns.
• New data shows just how wide the AI-hiring gender gap actually is.
• If you’ve ever wanted laundry detergent dropped aerially onto your front doorstep, Amazon announced its drone delivery program is expanding to almost 500 cities and towns by the end of the year.
• A new Comcast software upgrade turns existing Xfinity wi-fi routers into at-home motion detection sensors.
• AI may not be taking a cartographer’s job if this insane, AI-generated map of the US is any indication. What, you’ve never thought Illinois would be better renamed as “Vitoiis?”
• Breakdancing humanoid robots. Robots posing like Cruella de Vil’s dogs. Robots playing soccer. Here’s what’s going down at China’s 2026 World Robot Conference.
• Um, what exactly are Gwyneth Paltrow and Sam Altman doing together?

^{*A message from our sponsor.}

This time last week...

Readers’ most-clicked story was about Google’s answer to Apple’s AirTag—meet the Pixel Tag.