Three stories from the weekend that all circle the same question: who actually ends up holding the bag when AI money moves around. One lab is pledging a slice of itself to the public, one chatbot darling is getting picked apart in bankruptcy court, and one coding platform just changed horses in public.
01 of 03
Anthropic puts 5% of its shares into a public benefit fund
Anthropic announced Sunday that it is creating the Anthropic Public Benefit Fund, a wholly owned LLC, and seeding it with 5% of the company's shares, pending stockholder approval. On top of that, the board can drop in another 2% of shares each time the company's value climbs 50% above a set baseline. Anthropic has not said what that baseline is or whether there is a cap.
The stated purpose is dealing with AI's fallout, with labor market disruption at the top of the list. The fund can make large donations to nonprofits or send "direct distributions to specific populations." It is barred from using its equity for political donations or lobbying, in the US or anywhere else. Anthropic says more detail on who gets money is coming in the next few months.
Count in the founders' existing pledge to give away 80% of their equity plus an employee donation match, and Anthropic says roughly 20% of the company is earmarked for public benefit. The timing is not an accident. Anthropic is reportedly lining up an IPO as soon as November at a valuation north of $2 trillion, and it just told investors Q3 revenue grew 45% to $16.8 billion. At $2 trillion, that 5% is worth about $100 billion on paper. On paper is doing a lot of work in that sentence.
The critics showed up fast. Elon Musk said on X, "I recommend giving the shares directly to the public." David Sacks, who co-chairs the President's Council of Advisors on Science and Technology, wrote, "You've put 20% of your equity in an LLC. Whether it serves the public good remains to be seen." Anduril co-founder Palmer Luckey went further and called it "USAID 2.0."
Here is where I land. Sacks is annoying but he is right on the narrow point: an LLC the company wholly owns is not the same thing as giving stuff away, and we do not know who runs it. That said, I would rather a lab pre-commit stock to the people its product displaces than make the usual "we'll figure out UBI later" noises. The test is the first check. If the first big distribution goes to actual laid-off workers instead of a think tank, I will eat my words about the paper value.
Sources: Anthropic on X, Forbes, TechCrunch
02 of 03
Disney scoops up Character.AI's staff out of bankruptcy
Character.AI filed for Chapter 11 last week, with assets and liabilities both somewhere between $50 million and $100 million according to the court paperwork. Now a filing shows Disney agreed to offer jobs to at least 60 Character.AI employees, which is most of the company, at an estimated $60 million a year in total compensation. The filing says the "vast majority" who got offers took them.
That sits on top of an earlier deal where Disney agreed to buy some Character.AI assets for $20 million plus a non-exclusive perpetual license to its tech. It is a stalking horse bid, meaning it sets the floor and somebody else can still come in higher. The sale is expected to close in late November or early December. CEO Colin Adams has said that if no viable bid shows up, "the platform will be wound down."
The fall is steep. This is the company that hit 20 million monthly users by October 2024 and got a $2.7 billion licensing deal from Google that August, which also sent co-founders Noam Shazeer and Daniel De Freitas back to Google. It raised over $193 million from investors including Lightspeed, SV Angel, Greycroft, and a16z. The filing blames high costs and "potential litigation-related liabilities and regulatory engagement," which means the lawsuit from the mother of a 14-year-old who died by suicide after becoming obsessed with one of its bots, plus a child safety investigation from our own Texas attorney general. And there is a twist: former CEO Karandeep Anand just became Disney's first CTO.
What I keep coming back to is that Character.AI proved people would spend two hours a day talking to bots and still could not make the math work. Every extra minute of engagement cost real compute, and the users it was best at hooking were exactly the ones it got sued over. Disney walking off with the people for $20 million in assets is a bargain, and the guy who used to run the place is now on the buying side. That is not a conspiracy, but it sure is convenient.
Sources: court filing, TechCrunch on the Disney hires, TechCrunch on the bankruptcy
03 of 03
Replit makes OpenAI its default and says goodbye to Claude as the main engine
Replit CEO Amjad Masad said onstage at TechCrunch Disrupt on Thursday that Replit is no longer using Anthropic as its default model provider. His reasoning: "It was a decision based on both cost, partnership, and performance." He also said flat out, "I think ChatGPT is better now," pointing to Replit's internal and public benchmarks that he says favor OpenAI's Sol and Astra for building apps from scratch.
Over the weekend he cleaned it up a bit, telling TechCrunch that "Replit remains model agnostic" and that "OpenAI is now our preferred partner for powering Replit Agent." Claude is still available if users pick it. But most people never touch the default, so this is a real revenue hit. Masad said Replit was one of Anthropic's top five customers until recently. As recently as late September, a Replit blog post said Claude Fable 5 and Fable 5.1 were running Replit Agent in production. Replit announced its OpenAI partnership back in August. People familiar with the Anthropic relationship pushed back on Masad's version and one said the partnership is still in place. Anthropic did not comment in TechCrunch's report.
Masad also shared that enterprise went from about 2% of Replit's revenue last year to about 15% this year, and more than half of revenue is now B2B. He would not share ARR.
Here is my read. "Cost, partnership, and performance" is CEO for "they gave us a better deal." That is fine, it is a business. But it is a good reminder for anybody building on top of these models: you are one pricing sheet away from your vendor's biggest customer walking. Build your stuff so swapping models is a config change and not a rewrite, because the platforms you depend on are clearly doing the same thing to the labs.
Source: TechCrunch