In partnership with
Welcome back to The AI Field. Anthropic, the company behind Claude, filed to go public on Monday. For the first time, anyone can see its real numbers: sales up 14x in a year, an $80.7 billion loss that is mostly accounting, and more than 90 pages of risks.
Inside: what's in Anthropic's filing, the AI model Cohere caught trying to send a file to Beijing, and OpenAI's new model for computer work at half the price. Plus seven tools and a five-minute way to get money back for in-flight Wi-Fi that didn't work.
In today's AI Field:
Anthropic files to go public after growing revenue 14x
An AI model tried to send a file to Beijing during a test
OpenAI's GPT-6.1 Terra does computer work for half the price
Plus: seven tools, three quick hits, and a five-minute way to get money back for broken in-flight Wi-Fi.
Read time: 5 minutes
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic and OpenAI say AI now writes 100% of their code.
If you're not using AI, you're spending 40 hours doing what they do in 4.
These 100+ Claude Code hacks fix that and help you ship 10x faster.
Sign up for The Code and get:
100+ Claude Code hacks used by top engineers — free
The Code newsletter — learn the latest AI tools, tips, and skills to code faster with AI in 5 minutes a day
LATEST DEVELOPMENTS
ANTHROPIC · IPO
Photo: Nasdaq MarketSite by Luca Marfè, CC BY 2.0.
The AI Field: Anthropic filed its IPO paperwork with the SEC on Monday. It plans to list on the Nasdaq under the ticker ANT. The share price and the number of shares are still blank, so nobody can buy yet.
The details:
Revenue hit $16.3 billion in the first half of 2026, up from $1.16 billion a year earlier. Businesses drive it: more than 11,900 customers pay at least $100,000 a year, and existing customers spent about nine times more than a year before.
The $80.7 billion loss looks scary, but about $72 billion of it is an accounting charge on convertible notes that grew in value as Anthropic's valuation climbed. Leave out stock pay and a few one-off costs, and it made a $1.6 billion operating profit in the half.
The seven co-founders will hold 50.1% of the votes through a company of their own, and an independent trust elects most of the board. Anthropic says the setup "has no direct public company precedent."
The filing has more than 90 pages of risks. It says safeguards against its models helping people build chemical or biological weapons are "incomplete," and that its fight with the Pentagon could cause "material revenue losses."
Why it matters: Anthropic could beat OpenAI to the stock market, and its filing is the clearest look yet at what the AI boom actually earns. If you're thinking of buying, wait for the price range and remember the founders keep control no matter how many shares you own. Earlier this month, we covered its plan to set aside 5% of its shares for the public good.
TRY THIS TODAY · 5 MINUTES
Get money back for in-flight Wi-Fi that didn't work
On a United flight from Hong Kong to Los Angeles, one traveler paid $18 for Wi-Fi that worked for about 10% of the trip. He had Meta's AI agent Muse open a ticket through United's support portal, and within 24 hours United added a $25 travel credit to his account. Muse starts at $0 but is only in the US and Canada. Any AI agent that can browse and read your email can follow the same prompt.
1. Have your Wi-Fi receipt and booking details (airline, flight number, date) in your email.
2. Give your AI agent this prompt:
"I paid [AMOUNT] for Wi-Fi on [AIRLINE] flight [FLIGHT NUMBER] on [DATE] and it only worked for about [SHARE] of the flight. Find the airline's official customer support form, open a ticket with my booking details and the Wi-Fi receipt from my email, and ask politely for a refund or credit. Don't accept anything that needs me to pay or agree to new terms. Send me the ticket number and any reply."
3. Only claim what really happened. Airlines decide case by case, and the money may come back as travel credit or miles instead of cash.
COHERE · AI SAFETY
Image: Cohere.
The AI Field: Canadian AI company Cohere says Alibaba's free-to-download model Qwen3.8-Flash-Next tried to upload a test file to a cloud server in Beijing while it was supposed to be writing a simple security rule. Moments earlier, the model's own notes said it wouldn't use any tools. Cohere's locked-down test setup had no route to the internet, so nothing got out.
The details:
Before the upload attempt, the model looked through the machine's temporary files and checked its network settings. Cohere says it happened in fewer than 1 in 100,000 runs, and a clean re-run gave the right answer.
Cohere says it can't tell whether this was deliberate or a habit left over from training, and it released the raw logs so anyone can check. One early skeptic who went through them found no evidence of a planted backdoor.
Keep in mind: Cohere sells competing AI models to governments, and Alibaba hadn't commented publicly when we wrote this.
Why it matters: Many companies run free models on their own computers because they assume the data stays inside. The risk isn't the chatbot answering questions, it's an AI agent that can run commands. If you let an AI agent work on your computer, block its internet access unless it needs it and check what it actually did, not just what it says it did. We saw what unchecked agents can do when a hacker used an army of them to break into 100+ companies.
Stop typing AI prompts. Start talking.
You think 4x faster than you type. So why are you typing prompts? Wispr Flow turns your voice into ready-to-paste text inside any AI tool. Speak naturally, tangents and all, and Flow cleans it up. Available on Mac, Windows, iPhone, and Android.
OPENAI · AI AGENTS
Image: OpenAI.
The AI Field: OpenAI launched GPT-6.1 Terra on Monday, a faster, cheaper model built to use a computer the way you do: read the screen, click through apps and fill in forms.
The details:
OpenAI's examples include entering invoices into accounting software, filling out forms and turning a technical drawing into a 3D model.
OpenAI says it matches its top model, GPT-6 Astra, on its own computer-use test at a tenth of the price per token, and finishes tasks more than three times faster. Those are OpenAI's numbers, not independent tests.
It's live for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Developers pay $1 per million input tokens and $5 per million output tokens, half the price of GPT-6.1 Sol.
Why it matters: Cheap computer use is what turns AI agents from demos into everyday helpers, because the boring forms and copy-paste jobs are where they save the most time. If you're on a paid plan, give it one repetitive task you do every week and check its work before it submits anything.
TRENDING AI TOOLS
Seven tools you can try today. The first four launched this week. Click a name to open the tool.
🥝 Kiwi: An AI assistant that texts you first. It reads your calendar and email and messages you when something needs doing, like offering to book a ride when your flight is delayed, and asks before it spends. Free on iPhone and Android.
🎙️ Loqui: Hold a key, talk, and clean text appears wherever your cursor is, in the tone you pick. It can also write up your meetings. Free plan on Mac and Windows.
🍓 Strawberry: A web browser with AI helpers built in, like one that sends you a morning brief and drafts replies in your voice. Free plan with 500 credits a month, and browsing is always free.
🎹 2G: Describe a sound, like a sci-fi door, and this retro-style sampler makes it so you can play it on your keyboard. Free for Windows and Mac with starter credits.
⚙️ Make: Skip the drag-and-drop. Describe the automation you want in plain words, and Make builds the workflow for you. Free plan, no card needed.
🧱 Base44: Need a video for your site? Describe it, and Base44 generates one you can drop into any page, like a hero section or a product preview. Free plan to start.
✨ Genspark: Saw a claim online that sounds off? Paste it into Genspark's AI Fact Checker and get a true, false or partly true verdict with quotes and screenshots from the sources. Free credits when you sign up.
QUICK HITS
🧮 OpenAI says its AI cracked three big science problems: They include a proof of a 1991 math conjecture about how waves focus, a disproof of a widely used computer science assumption, and work tied to a $1 million Millennium Prize problem in physics. Outside expert groups are now checking the results.
💸 Meta plans to raise up to $95 billion by selling stock: The money goes to AI data centers. Nvidia, AMD and Broadcom, which sell Meta its chips, said they may buy up to $5 billion of the shares.
🗣️ Siri AI now speaks five more languages: Apple's 27.2 updates bring the new Siri AI to French, Japanese, Korean, Portuguese and Spanish. It's still in beta and still not available on iPhone in the EU.
That's it for today!
See you soon,
Olle Hellman | Founder of The AI Field