Tech Brew Ride Home - Anthropic Doing 2x OpenAI?

Episode Date: August 19, 2026

OpenAI paused training and rewrote its Preparedness Framework after Astra flirted with a critical cyber threshold. Leaked numbers showed Anthropic out-earning OpenAI nearly two to one, Anthropic prepped supervoting shares, Amazon expanded drone delivery, and Meta's addiction trial opened. Links OpenAI changed safety practices and paused RL training for two weeks after the Hugging Face breach and evidence Astra may have met a critical cyber threshold (Axios) Sources: OpenAI's Q2 sales grew 18% QoQ to $6.7B as its losses widened from $9.3B to $12.3B; Anthropic's sales grew 2x+ to $11.6B, with a small operating profit (The Wall Street Journal) Sources: Anthropic's revenue run rate reached $65B by the end of July, up from $47B in May 2026, $19B in March 2026, $9B in December 2025, and $4B in July 2025 (Bloomberg) Sources: Anthropic prepares to give its co-founders shares with extra voting power to help insulate them from outside pressure; Amodei owns ~2% of Anthropic (The Information) Amazon plans to expand Prime Air drone deliveries to cities in at least five more US states in the coming months, including Chicago and Atlanta by 2026's end (Bloomberg) In opening arguments, US state AGs say that Meta intentionally sought to addict children to Facebook and Instagram in pursuit of profit; Meta rejects the claims (Reuters) Subscribe to the ad-free feed.

Welcome to the TechBrewrite home for Wednesday, August 19th, 2026. I'm Brian McCullough today. OpenAI paused training and rewrote its preparedness framework after Astra flirted with a critical cyber threshold. Leaked numbers showed Anthropic out-earned Open AI nearly 2 to 1. Amazon expanded drone delivery and the social media addiction trial opened arguments. Here's what you miss today in the world of tech. IT pros can always be about fun and games until there's a fun game to actually play. Today's sponsor Eaton created an online car. card-based game called Ace of Uptime, and the whole idea behind it is to give IT pros a sandbox. You can practice problem-solving and decision-making and have some fun in the process. Each level of the game, you go up against a familiar foe like old processes or silent system failures, your Delta scenario, you choose your solution, and then see how your decisions play out. Will it resolve the problem?

Will it escalate it faster than expected? Start playing and find out. Head over to Eton.com slash ace to play Ace of Uptime. that's eton.com slash ace. OpenAI wants you to know it is continuing to pump the break, sort of, quoting Axios. OpenAI said Tuesday that it has made several changes to its safety practices following its determination that an upcoming system known as Astra may have reached a critical threshold for cybersecurity capabilities and the breach of hugging faces systems by another unreleased open

AI model. Open AI said Tuesday that it is in the process of rewriting its main security document known as the preparedness framework. Now that models are approaching or reaching the critical thresholds imagined in that document, most of which dates back to 2023. The company says it is also adding stronger monitoring across its development process, introducing alignment and security safeguards earlier in the development process and implementing higher safeguards than before when it scales up post-training. Open AI said it is increasing the compute resources devoted to understanding the underlying ways its systems are reasoning and acting. Open AI stressed that the new safety measures

are not simply a reaction to the hugging face breach, but part of a broader tightening of standards as models grow more capable. There is an incredible feeling of urgency to advance the levels of this sector and to prepare for the same kind of development happening outside of OpenAI and in the broader world. Chief scientist Jacob Pachaki said Tuesday during a briefing of reporters, end quote. And quoting Alex Heath's newsletter, I think it is a good time to slow down, OpenAI CEO Sam Altman told me last week describing the company's decision to halt the development of its most powerful unreleased models. The company announced Tuesday that it's implementing new safeguards that will slow its future AI development.

The company recently paused training on its next set of models, codenamed Astra for a little more than two weeks, according to executives, and its largest planned frontier training run remains on hold while the new guardrails are put in place. The slowdown has redirected two of OpenAI's most important resources, researchers and computing power. Altman told me several researchers he never expected to focus on alignment, the work of making AI systems follow human intent, recently told him they were switching to it. We've shifted a lot of compute, not just to alignment research,

but also to these new monitoring systems, he says. In exclusive interviews last week, Altman told sources and time that the slowdown was not caused by a single smoking gun, but rather a collection of research observations showing, quote, various degrees of misalignment as AI capabilities advanced faster than researchers had expected. The decision to tap the brakes could not be interpreted as evidence of any imminent catastrophe, he said. Getting AI safety right, Altman told me, is more important than any company's momentum. Open AI says it is now expanding safety monitoring across reinforcement,

learning, training, and evaluations, the phase of development in which advanced models are given the ability to use the internet and control software. The approach uses other AI systems to examine models' internal reasoning and behavior for unauthorized access, data theft, or attempts to defeat safeguards. In an interview, Pachaki told me that some of those protections go beyond OpenAI's current preparedness framework, its public rulebook for handling models that could cause severe harm. We don't have a date yet, he says, but we definitely believe we will need to evolve the preparedness framework. OpenAI said it plans to involve outside organizations as it revises the framework and will publish a detailed postmortem of the Hugging Face breach in the

coming days. Aside from the Hugging Face incident, OpenAI has not disclosed the results of its frontier research that led to the decision to slow down training, and the company leaders said the evidence behind Astra's risk classification will likely not be available until the model's technical report is released. We are very far from everything running back to normal. Mia Glaci, who leads safety and alignment work at OpenAIs, said Tuesday, end quote. Interesting that these details are leaking out now. Sources say OpenAIs Q2 sales grew 18% quarter over quarter to $6.7 billion, but its losses also widened from $9.3 billion to $12.3 billion.

Now, Anthropics sales reportedly grew 2x or more to $11.6 billion with a small operating profit. So Anthropic is doing almost twice as much revenue as OpenAI is right now, quoting the journal. The two companies diverging fortunes show just how drastically the artificial intelligence race has shifted since the beginning of the year. A slowdown in the growth of Chad GPT combined with the success of Anthropics hit coding tool Claude Code has put OpenAI on its back foot, forcing the company to pivot its business and overhaul its leadership team. Last week, the company replaced its chief revenue officer, Denise Dresser, after she spent less than a year on the job. Her departure followed a string of other exits, including former chief operating officer Brad Lightcap and Fiji Simo, once seen as the era parent to chief

executive Sam Altman. Open AI has told investors that its growth rate has picked up since the launch of a set of new models in July, the people said. For many startups, nearly $7 billion in quarterly would be extraordinary, but the expectations are different for OpenAI. The company has sold investors on a stratospheric pace of growth and signed large computing deals premised on its ability to soon generate hundreds of billions of dollars in revenue every year. The performance of Nvidia, Oracle, and other tech giants hinges on Open AI's ability to make good on its commitments to them. Open AI's sequential rate of growth was slower than Palantir over the same period, and also behind other AI high flyers such as Corweave and Micron, end quote. And quoting,

Bloomberg. Anthropic reported a preliminary revenue figure of more than $11.5 billion in its latest completed quarter compared with $787 million in the corresponding period in 2025, according to documents seen by Bloomberg News. It also reported positive adjusted operating income for the quarter, the documents show. Once viewed as the underdog, Anthropic has gained traction with AI tools that streamline complex tasks, including coding. Anthropics' revenue run rate topped $9 billion in late 2025 and crossed 47. billion in May. Anthropic now expects to generate annualized revenue of more than $65 billion based on its current performance, according to people familiar with the matter, up more than

sevenfold from its pace at the end of last year. The company's run rate, a metric that projects full year revenue from a shorter period, hit $65 billion by the end of July, said the people who spoke on condition of anonymity to discuss private information. Anthropic shared the figures as part of a regular update with investors, one person said, end quote. This is an interesting detail as we think about the forthcoming IPO, quoting the information. Anthropic has been preparing to give CEO Dario Amadai and other co-founders a class of stock with extra voting power to help insulate them from outside shareholder pressure to people familiar with

the matter said. It would be the first time Anthropics leaders have extra voting power, which has become common practice for tech founders. It is all the more relevant for Anthropic because its co-founders hold relatively small ownership stakes in the company compared to other tech founders, another person said. The company also is planning to maintain its existing body of non-shareholder trustees with a special class of stock to elect the majority of members to the company's board of directors, a more unconventional buffer against outside shareholder influence. Anthropic already has two structural components very few other large publicly traded companies have. It is a public benefit corporation, meaning its leaders have legal cover to pursue societal goals beyond making money for shareholders.

The IPO would make it by far the most valuable public benefit corporation in the U.S. stock market. The next largest is $40 billion software maker Viva Systems. And Anthropic has given significant but narrow governance powers to its long-term benefit trust, a group of advisors who aren't employees or investors. The trust created in 2023 has one major formal power today. It can elect the majority of Anthropic's seven-person board of directors. The company granted it those powers through class T shares that have no economic power. These trustees include former Federal Reserve Chair Ben Bernanke. The trust is down to just three members from its usual five after Mariano Florentino Quaylar, a former member of the California Supreme Court, exited this month to become Anthropics' chief global affairs officer. Anthropic is now planning to balance the trust powers with new voting rights for its seven founders. Because the company has raised hundreds of billions of outside capital, the founders combined stakes have been diluted down to a small minority ownership. Amadai himself owns only roughly 2% of the company a person close to

Anthropics said. That would be one of the smallest ownership percentages for any founder CEO taking his or her company public in recent decades. Amadai said on a podcast last year that he and the six other co-founders, including his sister, Anthropic President Daniela Amadai, each own roughly equal shares of the company. founders have often used special share classes to ensure control. Ahead of its IPO in June, SpaceX adopted a dual-class share structure that gave Elon Musk super voting shares, and other tech companies such as Snap and Meta platforms have also given their founders super voting shares. Meta CEO Mark Zuckerberg controls more than 50% of the voting

power, even though he has less than 15% of the company's overall shares, end quote. AI threats are like a hydra that learned how to type, cut one head down, and two more pop right back up, which means the old way we've been handling cybersecurity just won't cut it anymore. That's where Rubrik comes in. They've engineered a platform that secures your data, controls your AI, and protects your identity. See how Rubrik helps you stay ahead of threats at rubric.com slash mb. That's rubr-r-k.com slash mb. Are you a QuickBooks customer looking to grow your business without the growing pains?

You need the Intuit ERP. Upgrade to Intuit Enterprise Suite in a matter of hours. It's the AI Native ERP from the makers of QuickBooks. Learn more at Intuit.com slash ERP. Amazon plans to expand prime air drone deliveries to cities and at least five more U.S. states in the coming months, including Chicago and Atlanta by the end of the year. quoting Bloomberg. Prime Air drones will start delivering packages by the end of the year from warehouses in suburban Chicago and Atlanta, as well as Metro Cleveland and Syracuse, New York,

Amazon said in a statement on Wednesday. The e-commerce giant currently delivers from 11 U.S. locations, including the Houston and Dallas areas and suburban Phoenix. A company spokesperson declined to say how many of the new launch sites would be up and running by the end of the year, but said, the drone service would reach almost 500 cities and towns in the U.S., a six-fold increase from the current footprint. The company's MK30 drone designed and built-in house is a six-propeller craft that takes off and lands like a helicopter and has a range of about seven and a half miles. It has made hundreds of thousands of deliveries this year, a dramatic increase from the comparative trickle of the last few years when the company primarily made deliveries from since shuttered testing sites in California and Texas. Members of Amazon's Prime Service pay $2.99 for drone delivery with the fee waived on orders of $50 or more. customers without a prime membership pay a $499 delivery fee.

Amazon and its rival Walmart last year signaled big expansions for their drone delivery operations. Amazon has been retrofitting warehouses to accommodate its drones. Walmart, which flies drones built by partners including Alphabet's Wing, typically launches its craft from staging areas in store parking lots. Walmart's expansion late last year had it on track to offer airborne delivery to metro areas that are home to roughly one in every 10 people in the U.S. Wing in June said it was on track to fly from more than 270 Walmart locations within reach of more than 40 million people in the U.S. by next year.

Amazon is on the cusp of similar reach with plans to make drone deliveries available to 30 million people by the end of 2026, end quote. Finally today, reminder that this is going on in the background, quoting Reuters. Meta Platform's rejected accusations by U.S. states that it intentionally sought to addict children to its Facebook and Instagram platforms in pursuit of profit at the start of a trial that could reshape some of the most popular apps on the planet. A bipartisan group of 29 U.S. states is suing meta seeking potentially tens or hundreds of billions of dollars in penalties and changes to how meta does business. For lead states, California, Colorado, Kentucky, and New Jersey

accused meta of designing Facebook and Instagram to hook young users, feeling anxiety, depression, and even suicide, and misleading consumers about the platform's safety. All 29 states, States accuse META of violating federal law by improperly collecting and using children's personal data. Experts have called the trial in the Oakland, California federal court, which began with opening statements on Tuesday, the biggest legal test yet of social media's effects on young users. Megan O'Neill, a deputy attorney general for California, told the eight-person jury that Mata's business model was to, quote, hook the users, hold them for as long as they can, harvest their data and then hide the truth from the public. It worked especially well for kids, O'Neill added.

Meta needed kids and it needed to reassure the people who cared about those kids that the kids are safe. Meta's lawyer Paul Schmidt said there was no dispute that some social media users face struggles, but that research showed no clear link between adolescent social media use and a lack of well-being. He also said Mark Zuckerberg, Meta's co-founder and CEO, shared the company's desire to improve its services not to make them dangerous. They don't believe they're going to do well if people don't like their service, Schmidt said. jurors are expected to issue an advisory verdict, but U.S. District Judge Yvonne Gonzalez-Rogers will determine META's liability. If she finds META liable, Rogers could impose civil penalties and order changes to Facebook and Instagram.

META has said penalties could be as high as $1.4 trillion close to the Menlo Park-California-based company's market value. Attorneys General said at a hearing last week, the amount could be closer to $200 billion or about three years of after-tax profit for META. California, Colorado, Kentucky, and New Jersey also went meta to overhaul Facebook and Instagram, including by eliminating likes and the Infinite Scroll that encourages views of new posts, setting time limits for younger users and enforcing restrictions to keep children under 13 offline. After opening arguments concluded, former meta safety engineer Arturo Behar began testifying as the state's first witness. Behar has long said Meta knew its child safety tools did not work and has testified against the company in four trials. Mehta sought to block his testimony, allegedly for deleting signal messages with former employees,

but Rogers rejected what she considered a long-shot bid to eliminate a key witness. Behar told jurors that move fast and break things was a mantra at meta, which took a don't-ass, don't-tell approach to monitoring whether children under 13 were online. Many products were shipped into the world, such as Reels short-form videos and safety was not a consideration in how it was initially deployed, Behar said. Zuckerberg and Instagram chief Adam Osseri are also expected to testify, the trial is scheduled to last six weeks, end quote. Nothing more for you today. Talk to you tomorrow.