The week the maker of Claude showed investors $16.8 billion in quarterly revenue, a British doctor used Gemini to find the cause of his mother’s seizures, and Amazon, Delta and JPMorgan shut the door on AI agents from Meta, OpenAI and xAI.
- Anthropic reported third-quarter 2026 revenue of $16.8 billion, up 45% from the previous quarter, and its second consecutive adjusted profit.
- Anthropic plans a November stock market debut at a valuation of about $2 trillion, which would top SpaceX's record of $1.77 trillion.
- Amazon, Delta, JPMorgan and Yelp are blocking third-party AI agents from Meta, OpenAI and xAI, with Amazon having blocked Meta's Muse agent after 12 days, while Meta is working on a standard called PAP (Personal Agent Protocol).
AIPORT’s weekly roundup for October 12-18, 2026. Five minutes is enough to catch up.
Keep exploring artificial intelligence
Read the latest news and practical guides on AIPORT.
Anthropic’s $16.8 billion quarter sets up a November IPO at a roughly $2 trillion valuation
Anthropic, the AI company behind Claude, presented investors with results for the third quarter of 2026. Revenue reached $16.8 billion, 45% more than the previous quarter, and the company reported an adjusted profit for the second quarter in a row. The stock market debut is planned for November at a valuation of about $2 trillion, which would beat the record set by SpaceX ($1.77 trillion). In a separate interview, Dario Amodei said he spends as much as 40% of his time on company culture.
If Anthropic goes public at that valuation, it will be the largest IPO in history. It would also be further evidence that the market prices AI on real revenue rather than promises. For companies that use Claude, it means a supplier answerable to shareholders every quarter, which usually means steadier prices and fewer experiments.
Amazon, Delta, JPMorgan and Yelp are locking out AI agents from Meta, OpenAI and xAI
Large US companies have begun closing their doors to AI agents that would shop, book flights or manage accounts on users’ behalf. Amazon blocked Meta’s Muse agent after 12 days. Delta does not offer integrations. JPMorgan’s Jamie Dimon said bluntly that the bank does not want rogue agents on its systems. Schwab blocks such attempts dozens of times a week.
Each of these companies is promoting an agent of its own: Amazon has Rufus, Delta has Charley and JPMorgan has Smart Cash. Meta is working on a standard called PAP (Personal Agent Protocol) to bring order to the situation.
The dispute is one of the biggest of the fall. The vision of telling ChatGPT to buy a plane ticket and having it handled on any website runs into the fact that retailers and airlines do not want to hand the customer relationship to someone else. For the coming months, AI agents will mostly work where the service owner built them itself.
A doctor used Gemini to trace his mother’s seizures to a vitamin B6 deficiency
British anesthesiologist Ed Patrick used Google Gemini to review medical publications while his mother was dying in the hospital from seizures that doctors could not control. The model pointed to a vitamin B6 deficiency caused by levodopa, a Parkinson’s disease drug. After she was given the vitamin, the seizures stopped within minutes, and a neurologist confirmed the lead was sound. The case was reported by the BBC and The Sunday Times. The US FDA had previously ordered changes to the drug’s labeling over this risk.
The story is not about AI replacing doctors. Patrick is a physician, so he knew how to ask the question and how to judge the answer. It does show that a language model can search, in minutes, literature that a busy hospital team had no time to read.
Tell our editorial team about your project. Include a short description, the relevant links and your contact information.
Character.AI declares bankruptcy, and Disney pays $20 million for a license, not the platform
Character.AI, one of the most popular apps for chatting with AI characters, declared bankruptcy. Disney will pay $20 million, but only for a license, not to acquire the platform. The same day, Epic Games said it was delaying a shooter it is building with Disney and moving its production to an “AI first” model.
Character.AI’s collapse is the first this prominent for a company that two years ago was seen as one of generative AI’s stars. It shows that millions of users are not enough without a way to make money, especially as the cost of serving each conversation rises. There is a darker side for parents too: an app used by a great many teenagers is ending amid lawsuits and child-safety problems.
OpenAI fired three safety researchers, including its liaison to METR
OpenAI dismissed three AI safety researchers, Tomek Korbak, Wang and Balesni. One of them was responsible for contact with METR, an organization that runs independent tests of AI models. The fired researchers published an open letter questioning the voluntary audit model on which the whole industry now relies. A few days earlier, the FBI detained OpenAI engineer Abdulrahman Alfozan, who allegedly hid a listening device in the company’s boardroom.
The case matters more than it looks. Oversight of the most powerful models rests on companies inviting outside auditors themselves. If the people who built that bridge are fired and say publicly that it does not work, lawmakers in Brussels and Washington get a ready-made argument for tougher rules. Bill Gates, the same week, announced a personal campaign for AI regulation.
Meta’s Muse Wonder ties the top tier, and Sam Altman confirms an OpenAI humanoid robot
Meta released the Muse Wonder family of models. The Wonder Full version matched Claude Sonnet 5.5 in the Artificial Analysis ranking, ahead of GPT-6 Astra and two points behind Claude Opus 5.5. Sam Altman confirmed that OpenAI will build its own humanoid robot, with a first demonstration planned for 2027.
The same week, Google made Gemini 4 Argon available to paying users (API access through November 30 at half the target price). China’s DeepSeek showed a model, V4.1-Flash-1017, that beats Claude Opus 5 in agentic tests.
For users, the differences between the best models have become cosmetic. Meta, which a year ago trailed the pack, now sits at the same table as Anthropic, OpenAI and Google. OpenAI’s robot signals that the race is moving from the screen to the physical world, where China has the edge: Beijing-based LinkerBot controls more than 80% of the robotic hands market.
Bezos promises a three-day week while Kroger cuts 3,000 jobs, and AI insiders expect far more change than the public
Jeff Bezos said AI will let some people support a family while working three days a week. The same week, grocery chain Kroger announced it will lay off 3,000 headquarters employees over three years, and CEO Greg Foran explained the cuts as a bet on AI.
A Financial Times survey of 94 economists found that most expect AI to speed up economic growth outside the US and China, but without mass unemployment. Anthropic measured what it calls an “expectations gap”: the company’s employees think AI will do 75% of their work by 2030, while the average American says 28%.
The narratives are pulling apart. A billionaire promises a three-day week, a retailer cuts jobs, economists reassure, and people in the AI industry expect much deeper changes than the rest of society. Who is right will be clear in a few years. For now, one thing is certain: the expectations gap is also a preparedness gap.
Blunder of the week: Swedish broadcaster TV4 delays a finished season, and an AI agent may have deleted it
TV4, a Swedish broadcaster, pushed the premiere of a finished season of the reality show The Challenge Sverige to 2027. According to reports, the footage was deleted by an AI agent that was tidying up resources. The station officially cites human error. Commentators note that this does not rule out the agent version. It only shifts responsibility to the person who allowed the agent to do it.
Runner-up in the category is Eiliko, an AI plush toy from Energize Lab that stored 13 million children’s conversations on Chinese servers. A security researcher remotely took over a unit belonging to a journalist at The Verge.
Claude can work as a demanding tutor with nothing but a good prompt
No new app is needed. The AIPORT editorial team tested six prompts that turn Claude into a tutor that does not hand over answers. Instead, it asks questions, checks understanding and returns to a topic when you skip something. It works on a free account, in a browser.
How to try it tonight:
- Go to claude.ai and create a free account if you do not have one.
- Write a prompt along these lines: tell Claude to act as your tutor in a chosen subject, to withhold answers and to guide you with questions.
- Paste in the topic you are struggling with: a child’s math homework, English before a job interview, rules and regulations you do not understand.
- Answer the questions instead of asking for a ready-made solution. After 20 minutes you will see the difference.
The week in one line
Anthropic is counting billions and Bezos is promising a three-day week. See you next week.
Sources and methodology
The original article does not link primary documents. The reporting it relies on, as named in the text:
- Anthropic’s third-quarter 2026 results and IPO plans, including Dario Amodei’s interview on company culture (no primary link in the source)
- BBC and The Sunday Times, on the Ed Patrick case (no URLs in the source)
- Artificial Analysis ranking, on Meta’s Muse Wonder (no URL in the source)
- Financial Times survey of 94 economists (no URL in the source)
- The Verge, on the Eiliko toy (no URL in the source)
Analysis and editorial responsibility: Piotr Wolniewicz, Editor-in-Chief, AIPORT. Original Polish reporting: Piotr Wolniewicz (aiport.pl).
This article was originally published in Polish on aiport.pl on October 19, 2026.
https://aiport.pl/co-nowego-w-ai/ai-dla-zajetych-43-12-18-pazdziernika-26-anthropic-liczy-miliardy-przed-gielda-gemini-ratuje-zycie-a-zus-przeswietla-l4/
