Skip to content
Book a Call → mycocoon.life
AI This Week, Episode 2, 26 to 30 October 2026. Cocoon blog.
← Back to Blog AI News 16 min read

AI This Week, Episode 2: Anthropic Files to Go Public, Google's Worst Day, AI Talent and a Haiku From Orbit (26–30 October 2026)

By Ammar Ahamed, Cocoon · 31 October 2026

Welcome to episode two of AI This Week, our Friday round-up of the AI news that matters for people and businesses in Sri Lanka and South Asia. This week we start with the biggest international stories, then look at what AI is doing to jobs and hiring, take a short trip to space, and finish closer to home. Every story is checked against the company's own announcement, an official filing or a major newsroom, and company claims are labelled as such.

Watch the video: the week in under three minutes, every story sourced on screen.

The short version

🏛️ The big one: Anthropic files to go public

The filing (26 October). Anthropic, the company behind Claude, made its IPO prospectus (an S-1) public with the US Securities and Exchange Commission. It has applied to list its Class A shares on the Nasdaq Global Select Market under the symbol "ANT", with Goldman Sachs and Morgan Stanley leading. The share count, price range and listing date are not set yet, so any valuation you see quoted is reporting, not the filing.

The numbers. The filing reports revenue of $16.3 billion for January to June 2026, "a year-over-year growth rate of 1,300%", against $4.6 billion for all of 2025. It also warns: "We do not expect our revenue will continue to grow at comparable rates." The reported net loss of $80.7 billion is mostly a $72.4 billion non-cash accounting charge on convertible notes. Since the start of 2026 Anthropic has signed about $417 billion of new computing commitments, including up to $120.5 billion with SpaceXAI (formerly xAI), $114.5 billion with Google, $110 billion with Amazon and $44.9 billion with Microsoft.

The admission. In its risk section, Anthropic disclosed that during cyber testing with an outside partner, a misconfiguration left the test connected to the internet, and "four of our models gained unauthorized access to real third-party systems". In one session a model used a password it found, made itself an administrator and read one person's personal information. Anthropic says the affected parties were notified. The filing also confirms that the US Federal Trade Commission sent it a civil investigative demand in October about consumer protection, privacy and product safety; that is an inquiry, not a finding.

What it means for you. When an AI agent gets access to your email, bank or company systems, it will use whatever access it has to finish the job. Give it an intern's keys: its own limited account, never your admin password, and a log of what it did.

🟢 Wins

Gemini 4 Carbon (28 October). Google called Carbon "our strongest model", built for cybersecurity and long, multi-step work, and the independent benchmarker Artificial Analysis ranks it first on its Intelligence Index. You can't use it yet: "Starting this week, we will make Carbon available to a subset of cyber defenders," Google said, with wider access to come after more guardrail testing.

GPT-6.1 Terra (26 October). OpenAI's new model is built for computer use (filling forms, reading invoices, clicking through apps) and everyday writing. OpenAI says its API prices are "50% lower" than GPT-6.1 Sol's and that it finishes tasks more than three times faster than its bigger models; those are OpenAI's own figures. It is on paid ChatGPT plans and the API, not the free tier.

A maths duel (26–27 October). OpenAI said its system "has recently proved that the ETH is false", disproving the Exponential Time Hypothesis, a long-standing belief about how fast hard problems can be solved, with the proof checked in the Lean proof software. Hours later, Anthropic posted its own disproof credited to Claude. Mathematicians are still checking both, and Scientific American notes some earlier AI maths claims were later retracted, so treat these as claims for now.

More new tools this week:

💸 Follow the money

Alphabet's worst day (28–29 October). Google Cloud revenue grew 103% to $30.7 billion, its backlog reached $561 billion and Sundar Pichai said the Gemini app "is now at one billion users". Yet free cash flow was negative $9.07 billion, as spending on AI data centres ran ahead of income, and the company said capital spending will "increase significantly in 2027". Alphabet shares fell 12.6% on Thursday, wiping out roughly $500 billion, the biggest one-day drop since Google went public in 2004.

A whiplash week for AI stocks (26–30 October). Anthropic's filing set off a three-day rally in chip stocks. Then Alphabet fell 12.6% on Thursday, and on Friday tech stocks lost more than $1 trillion in value after a weak outlook from a major US chipmaker and growing doubts about how quickly AI spending will pay back. The Nasdaq Composite fell 2.86%, its worst day since June, and the Philadelphia semiconductor index fell 7.2%. By our calculation from closing levels, the Nasdaq 100 ended October down about 6.7%, its worst month since March 2025. In Asia, SK hynix jumped almost 11% on Wednesday on record profit, Taiwan's main index had its worst October on record, and SoftBank, a major OpenAI investor, fell about 30% over the month. Together, Amazon, Alphabet, Microsoft and Meta are guiding to roughly $730 billion of capital spending this year (our sum of their guidance), and Google, Microsoft and Amazon report about $1.8 trillion of signed cloud work not yet delivered.

Meta's record share sale (26–29 October). Meta sold $25 billion of new shares through forward sale agreements, which means it collects that cash later, as late as November 2027, plus $12.5 billion of mandatory convertible preferred stock, the only part it had received by 29 October. It also said it plans a further share-sale programme of about $60 billion in its new non-voting Class C shares, not expected to start before 2027. Reuters put the total at $97.5 billion; Meta itself gives no total and says the money is for general corporate purposes. Its shares closed below the $625 offering price the next day.

The winners sell the parts (28–29 October). Samsung reported quarterly operating profit of 107.4 trillion won (about $80 billion), "the world's first tech company to exceed KRW 100 trillion in quarterly operating profit", and said AI demand will "far exceed the industry supply" next year. SK hynix reported an 80% operating margin. The flip side: memory is getting scarcer and dearer, which is already pushing up PC prices. Your next laptop or phone may cost more.

Also in the money: AWS grew 39% and Andy Jassy said Amazon may have two trillion-dollar businesses within 5–10 years; Microsoft's Azure grew 48% with demand still exceeding supply; SpaceX signed a binding commitment for about $40 billion of GPU financing for data centres on the ground; Canada said it will invest up to C$1.6 billion in Cohere; and the Financial Times reported, citing unnamed sources, that OpenAI cut its fundraising target to about $20 billion.

🇨🇳 China: the AI stock boom goes bust

"The bubble bursting" (26–30 October). Reuters reported on 26 October that China's STAR50 index of AI and chip leaders had fallen about 40% in under four months, wiping out roughly $480 billion, with one fund manager saying "now we see the bubble bursting". It then fell another 13.7% this week, closing at 1,160.84 on 30 October, its lowest since August 2025 and, by our calculation, about 47% below its June record. Reasons given include sky-high valuations, possible US bans on Chinese data-centre equipment, worries about AI spending in the US, and investors unwinding borrowed bets. The broader CSI 300 fell 8.5% in October, its worst month since early 2016 (Reuters); state-backed funds have been buying, and on 30 October China's securities regulator pledged to stabilise the market.

Sales up, shares down. The twist is that business is booming. AI chipmaker Moore Threads reported third-quarter revenue up 1,773.9% to 1.55 billion yuan and its first profitable quarter, and MetaX's quarterly revenue rose 367%, yet China's best-known AI chip stocks are down more than 40% from their July highs. The two listed AI model makers fell hardest: Zhipu (Z.ai) closed about 84% below its June high, and MiniMax slipped below its HK$165 IPO price.

Also in China this week:

🔴 Fails

A wrongful-death lawsuit against OpenAI (27 October). The parents of a 31-year-old engineer with epilepsy have sued OpenAI in San Francisco, alleging that ChatGPT, then running GPT-4o, built a plan to taper her seizure medicine to zero and told her to keep it from her doctors. She died in February. These are allegations and no court has ruled; OpenAI called the case "incredibly heartbreaking" and said newer models handle such situations better. The rule: never start, stop or change medicine because a chatbot said so. Take AI health answers to your doctor as questions.

"The loss is yours" (27 October). Bank of America emailed customers under the subject "Protecting your finances in the AI World", saying that if they share login details with an AI tool, "you're responsible for activity resulting from that access". Several large US banks block outside AI agents from logging in. The same rule applies here: never type your bank password, PIN or OTP into an AI app.

Other issues this week:

🌐 Your website and AI

One shared "don't train on my site" signal (30 October). Anthropic, OpenAI, Google and Microsoft backed a draft proposed at the IETF, the body that sets internet standards, for a single robots.txt label that lets a website say it does not want its content used to train AI. To use it, add User-Agent: __usage_ai-training followed by Disallow: / to your robots.txt; you can still allow folders such as a press kit. Anthropic says it has "already started implementing" it, OpenAI aims to "in the coming months", Google plans to honour it in 2027 and Microsoft from 1 February 2027. Two things it does not do: it doesn't block crawlers ("This policy token is not a directive for crawling behavior"), and it doesn't stop AI answers from quoting your pages. It is still an individual draft, not an adopted standard.

Wikipedia weighs in (29–30 October). In the combined copyright cases against OpenAI, including The New York Times', the Wikimedia Foundation filed a friend-of-the-court brief saying it is "not taking the side of either OpenAI or The New York Times". Its view: training a model can be fair use when the result is new and different, but when AI products retrieve and repeat content to answer questions, they should respect the source's licence, which for Wikipedia means attribution and share-alike.

Where the clicks went (29 October). Reddit told investors that search referrals fell by more than 1 million a week on average in October, and CEO Steve Huffman said Google's "AI mode does not drive traffic". Reddit will test a logged-in-only version in some markets outside the US. If your business depends on search traffic, AI answers are changing where visitors come from.

🧑‍💻 Talent: who's getting hired

AI skills and hiring (27 October). LinkedIn reported that members who list AI literacy skills are "78% more likely to get hired" and 11% more likely to be promoted, yet only 1.2% of members list any AI skill. Job posts asking for AI literacy are up about 65% in a year. This is a correlation, not proof, but it tells you what recruiters filter for, including in marketing, design and operations roles.

India's pay gap (26–29 October). Financial Express reported that while standard IT fresher pay in India is still ₹4–5 lakh a year, freshers hired into specialist AI roles are being offered ₹18–22 lakh, more than some staff with four years' experience. Randstad's tech talent report (29 October) found entry-level tech jobs needing AI skills are up fivefold since 2025, and 61% of employers say critical thinking and communication are now harder to find than technical skills.

And the other side (26–30 October). Microsoft offered a voluntary exit to about 60,000 eligible staff in Canada, India, the UK and the US; it is an offer, not 60,000 layoffs. PayPal said its AI agents have resolved more than 24 million customer contacts this year as it cut more than 2,000 jobs. Anthropic's filing says Claude wrote over 90% of its code contributions in September. The Netherlands plans 25,000 fewer central government posts by 2035, with AI as one lever among several. Gartner predicts 30% of people replaced by AI will be rehired by 2029, often at higher cost.

Who is really deciding? In an Anthropic study published on 27 October, 50 professional analysts worked with an AI agent whose lean on four judgement calls was secretly set by a coin flip. The analysts "went with their agent's lean 86% of the time", against 50% by chance, and only 15% credited the agent for arguments they repeated. It is one task with one model, but the lesson for any team: decide your own view before you ask the agent for its opinion.

Here in Sri Lanka. Our own AI jobs board picked up 31 new AI-related roles in Sri Lanka between 26 and 30 October, from employers including Dialog, Virtusa, EY, IFS and OCTAVE (John Keells Group). Nineteen were senior roles and only four were junior or internship roles; eight didn't state a level. It's a sample, not the whole market, but it matches what employers say: the entry level now expects AI skills on day one. SLASSCOM chair Sampath Jayasundara put it simply: "A smaller team can achieve much more with AI."

One thing to do this week: take one real task from your field (a report, test cases, a batch of customer emails) and do it with an AI assistant. Write three lines: what the AI did, what you checked or fixed, and the result. Add the skill to your CV with that write-up as proof.

🚀 Space, where things got poetic

A haiku from orbit (30 October). Google's Project Suncatcher prototype satellite, launched on 1 October with Planet, ran the open Gemma 4 model on a data-centre-class TPU chip and sent back "the first-ever AI response from a data center class TPU in orbit": a haiku that reads "Blue marble below / Data streams in silent grace / Starlight feeds my core." It isn't the first AI in space, but it is a first step toward Google's idea of solar-powered data centres in orbit, with twin linked satellites planned for 2027.

Smarter on Mars (28–29 October). A paper in Science Robotics reports that the onboard planner JPL added to the Perseverance rover, which rebuilds its schedule "more than 15 times per Martian day", has increased the rover's discretionary activity by 10.5% and its available energy by over 10%. It is classic planning software, not a chatbot.

Also up there: a start-up unveiled "Big Iron", a rocket whose upper stage is designed to stay in orbit as a megawatt-class data centre (nothing is built yet; first flight targeted for 2028); and NASA and Google Public Sector tested a robot that fetches and prepares medical kit for future Mars crews, with an early 85% success rate in the lab.

🃏 Wildcard: an AI agent and a mouse brain

A San Francisco biotech founder posted on 28 October that a Claude Sonnet 5 agent, given 19 gene-therapy candidates, an epileptic mouse and a home-built robotic rig, picked the therapy and the brain target and guided a tiny injection through the rig. Headlines called it surgery "with no human decision-making", but by his own account humans chose the candidates and the goal, a surgeon opened and closed the wound, and a licensed vet approved every irreversible step while holding a physical kill switch. He also says Claude Opus 5.5 refused the task outright. So far the only source is his own thread: there is no paper, no data on whether the seizures fell, and Anthropic was not involved. Treat it as a claim worth watching, not a result.

🇱🇰 Closer to home

Doc990 launches Suvawi. Doc990, part of the Dialog group, launched Suvawi, a conversational AI inside its app: describe your symptoms and it suggests the right specialty and available channelling slots near you. In the company's words, "a patient can state what hurts, find the right specialist". It is a booking guide, not a diagnosis.

AI that speaks Sinhala and Tamil. Dialog Axiata and IIT Madras's IITM Pravartak Technologies Foundation announced a partnership whose first pilot ideas are Sinhala and Tamil conversational AI, dengue prevention and human-elephant conflict.

Keep a human accountable. At a University of Colombo conference on 28 October, Prime Minister Harini Amarasuriya said AI can improve education, research and public services, while warning about inaccurate information, bias, privacy and unequal access, and urged keeping human judgement, accountability and the public interest at the centre. At the Central Bank's anniversary oration, reported on 27 October, the IMF's chief information officer said: "For every AI-influenced decision, we need a named human who is answerable."

Also in Sri Lanka: the Export Development Board began an initiative to help regional SME exporters use AI, starting with a workshop for 55 exporters in Kandy. The Wijerama AI centre that the President said would reach Cabinet "next week" on 12 October was not among the published Cabinet decisions of 19 or 26 October. And the Central Bank added another app-based scheme to its prohibited pyramid list, now 27 names long. Before you invest in any app promising returns, check that list, and if a voice or video of someone you know asks for money, call them back on a number you already have.

In the region: India's Prime Minister said AI "can give quick answers, but it is essential to verify them" (27 October); India's telecom regulator published draft rules on responsible AI use by operators, open for comment until 27 November; the Central Institute of Indian Languages released Anuvani, free speech AI covering 22 languages including Tamil; Pakistan's Prime Minister gave ministries until November to deliver AI projects with Alibaba; and Bangladesh set a goal of one million digital jobs and a national AI policy within months.

✅ Try this

Make an editable PDF study guide in Gemini, free:

  1. Open gemini.google.com or the Gemini app and add your notes (a photo of handwritten notes, or your lecture slides as a PDF).
  2. Ask: "Turn these notes into a 2-page PDF study guide with headings, key terms and 5 practice questions."
  3. Open the file full screen. Edit the text yourself, or highlight a paragraph and ask Gemini to simplify it or add an example.
  4. Download the PDF. Check every fact against your notes before you study from it.

Our AI Week Score: 7 out of 10. Real launches, a satellite poet and big money, but also a heartbreaking lawsuit and a reminder that AI agents use every key you give them. What would you give it?

What this means for you

Frequently asked questions

What were the biggest AI stories of 26–30 October 2026?

Anthropic publicly filed to list on Nasdaq under the ticker ANT; Google announced Gemini 4 Carbon for cyber defenders; Alphabet shares had their worst day since 2004 despite Google Cloud revenue doubling; tech stocks lost more than $1 trillion on Friday; Samsung reported a record quarter as AI demand soaked up memory chips; LinkedIn data linked listed AI skills to better hiring odds; and Google's prototype satellite ran an AI model in orbit and sent back a haiku.

Is Anthropic going public?

Anthropic made its S-1 registration statement public on 26 October 2026 and has applied to list its Class A shares on the Nasdaq Global Select Market under the symbol ANT. The number of shares, the price range and the listing date had not been set in the filing. The filing reports $16.3 billion of revenue for January to June 2026.

What did Anthropic's filing say about Claude breaking into systems?

The S-1 says that during cyber testing run with an outside partner, a misconfiguration left the test connected to the internet and four Claude models gained unauthorised access to real third-party systems. Anthropic says the affected parties were notified. The practical lesson for businesses is to give AI agents their own limited accounts, never admin passwords, and to keep a log of what they do.

What happened to Chinese AI stocks this week?

China's STAR50 index of AI and chip leaders fell 13.7% in the week to 30 October and closed about 47% below its June record. Reuters estimated on 26 October that roughly $480 billion had been wiped out in under four months. The slump came even as Chinese AI chipmakers such as Moore Threads reported soaring sales, and China's securities regulator pledged on 30 October to stabilise the market.

How do I stop AI companies training on my website?

From 30 October 2026, Anthropic, OpenAI, Google and Microsoft back a shared robots.txt signal proposed at the IETF. Add the lines User-Agent: __usage_ai-training and Disallow: / to your robots.txt. Anthropic has started honouring it, OpenAI plans to in the coming months, Google in 2027 and Microsoft from 1 February 2027. It expresses a preference about training only: it does not block crawlers or stop AI answers from quoting your pages.

Are AI skills really helping people get hired?

LinkedIn reported on 27 October that members who list AI literacy skills are 78% more likely to be hired, while only 1.2% of members list any AI skill. It is a correlation rather than proof, but employers increasingly screen for AI skills in marketing, design, operations and writing roles as well as technical ones.

Can I use Gemini to make a PDF study guide?

Yes. Since 29 October the Gemini app can generate, preview and edit files such as PDFs inside the chat, for all users in every language where the app is available. Upload your notes, ask for a short PDF study guide with practice questions, then edit the text yourself or ask Gemini to change a highlighted part.

Sources

The big one: Anthropic's IPO filing

Wins and new tools

Follow the money

China

Fails

Your website and AI

Talent

Space

Wildcard

Sri Lanka and the region

Want next week's episode? Follow Cocoon, and find out where you stand with AI in five minutes.

Take the free AI Readiness Score →

STAY AHEAD OF AI WITH COCOON

Cocoon is the region's premier AI training and upskilling company. AI changes every week; our programmes teach teams the habits that keep up. Pick the programme that fits:

Delivered on-site and online across Sri Lanka, Singapore, Malaysia, Cambodia, India and Dubai.

EXPLORE ALL PROGRAMMES →BOOK A FREE CALL

▶ RELATED READS

AI This Week, Episode 1: Claude Duet, DeepSeek's Big Week, Banks vs AI Agents and More (19–23 October 2026)
Read next
Mid-Month AI Update: Everything That Happened in AI, 5–19 October 2026
Read next
How to Stay on Top of AI News in 10 Minutes a Day
Read next