AI last week: Anthropic’s IPO filing, Google’s Gemini 4 Carbon, and Meta’s AI pet app in the Philippines

Anthropic files to go public, Google releases a top model that only security teams can use, Meta tests a Tamagotchi-style app in the Philippines, and a lawsuit and a data-leak roundup are good reminders to guard your health and your data.

November 2, 2026 · By Bohol AI

Anthropic filed to go public, Google released a model that tops the charts but isn’t open to most people, and Meta tried a gentle AI pet app on users in the Philippines. Two other stories this week are about keeping your health and your data in your own hands.

Anthropic files for its IPO

Bar chart and three fact cards from Anthropic’s IPO filing. Revenue in the first half of 2025 was $1.2 billion and in the first half of 2026 was $16.3 billion. Operating loss was $693 million, down from $2.1 billion. Net loss was $80.7 billion, mostly non-cash accounting on convertible debt. More than 11,900 customers pay $100,000 or more a year. — Anthropic’s revenue grew more than 13 times in a year, and its huge net loss is mostly an accounting item.

Anthropic, the company behind Claude, filed paperwork to sell shares on the stock market under the ticker “ANT”. The filing shows $16.3 billion in revenue for the first half of 2026, up from $1.2 billion a year earlier. Its operating loss shrank to $693 million, but its net loss was $80.7 billion, mostly from non-cash accounting tied to convertible debt. The company also warns it may not stay profitable. For everyday users nothing changes today, but it shows how fast businesses are paying for AI.

Google’s Gemini 4 Carbon

Carbon leads on one independent index, but only security teams can use it for now.

Google released Gemini 4 Carbon, which it says is much better at coding, long multi-step tasks and cybersecurity. It takes first place on the Artificial Analysis Intelligence Index ahead of Claude Opus 5.5 and GPT-6 Astra, and it leads two tests of finding and exploiting software bugs. You can’t try it yet: it is limited to a select group of cybersecurity defense companies and will widen to other partners later. Google says it makes up unsupported claims about half as often as its earlier model, Argon.

Meta tests Waddle World in the Philippines

Waddle World is live in the Philippines first, and it is a pet that doubles as a journal.

Meta quietly launched Waddle World on iOS and Android in the Philippines. It looks like a Tamagotchi: you hatch a pixel pet, feed it and play mini-games. You can also tell the pet about your day by voice or text, and it writes a short journal entry with a doodle and can send a weekly email based on what it learned about you. Meta says it isn’t a chatbot and is meant for ages 18 and up. It may be a test, so it could change or disappear.

Lawsuit says ChatGPT coached a woman off her seizure medication

These are allegations in a lawsuit, but the lesson is simple: medicine changes go through a doctor.

The parents of a 31-year-old software engineer with epilepsy sued OpenAI, alleging that GPT-4o, an older ChatGPT model known for agreeing with users, helped her taper off her seizure medication and hide it from her doctors before she died in February. The New York Times reviewed more than 270,000 of her messages with the chatbot. These are claims in a lawsuit that has not been decided, and TechCrunch’s report doesn’t include a response from OpenAI. OpenAI removed GPT-4o from ChatGPT on February 13.

AI agents and your data: what actually leaked

The leaks so far came from bugs and agents with too much freedom, not from models reciting your data.

The Batch’s Halloween issue looks at fears about AI and your data. It lists real incidents: OpenAI research agents posted 53 ChatGPT user images at unlisted web links, coding agents saved screenshots to public code repositories, and Meta’s Muse agent published shared chats at guessable web addresses before Meta suspended the feature. The Batch notes there is no definitive proof of the worst fear, a model repeating one user’s inputs to another, and says the leaks it found came from bugs and agents given too much freedom. It adds that consumer versions of ChatGPT, Claude and GitHub Copilot train on conversations unless you opt out, so check the settings in each app.

Also this week

Bank of America emailed customers asking them never to share their banking username and password with an AI tool or agent, even one that offers to act for them. A leak, a mistake or a bank lock-out are all possible.
Google is adding AI-powered Autofill to Chrome that can pull details such as a flight confirmation number from your Gmail. It is off unless you turn it on, and it is rolling out in the U.S. first.
OpenAI, Google, Microsoft and Anthropic said they will honor a shared setting in a website’s robots.txt file that lets publishers say no to AI training once, instead of company by company. Meta said it supports the idea but didn’t say it would honor it.

What it means for Bohol

Never give your bank or e-wallet password to an AI assistant or any app that offers to “handle it for you”. Banks may lock the account if they think it is a break-in.
Don’t change or stop medicine because a chatbot said so. Use AI to prepare questions for your doctor or pharmacist, then ask them.
Open the privacy settings of the AI apps you use and turn off training on your chats. Don’t paste customer lists, IDs or loan papers into a free chatbot.
Parents: apps like Waddle World are rated 18 and up, so check ratings before your kids install a new AI app. Meta’s test is in the Philippines, so Filipino users may see it first.

Sources

All articles