Briefing #10 10 min read

GPT-6 arrives in ChatGPT and now builds calculators right in the chat

OpenAI switched all of ChatGPT to GPT-6, free accounts included. Also: 77 zero-days in two days of Pwn2Own, Windows 11 fences in AI agents and Waymo borrows 5 billion dollars.

Black-and-white 3D render illustrating: OpenAI switches ChatGPT to GPT-6 and debuts answers with charts and calculators
Editorial 3D render made for this edition.

TL;DR Quick summary

  • On October 7 OpenAI started switching ChatGPT to GPT-6: Sol for paid plans and Luna for Free and Go, the following day.
  • The visible change is Intelligent UI: answers with charts, buttons, forms and even calculators you can adjust yourself.
  • In its safety card, OpenAI reports regressions for both models in its tests for users under 18.
  • Two days of Pwn2Own Ireland produced 77 zero-days; the Galaxy S26 fell six times.
  • Windows 11 now ships containers to limit AI agents, and Waymo closed its first loan: 5 billion dollars.

The briefing on video

Temporary narration with a synthetic voice.Vertical version 9:16

Video transcript

ChatGPT got a new brain this week, free accounts included, and instead of explaining the math to you, it now builds you the calculator.

On October seventh OpenAI started moving ChatGPT to GPT six. Sol for paying users and Luna for free accounts. The visible change is called Intelligent UI: answers with charts, buttons, forms and calculators you can adjust yourself.

Before, it was like asking the guy at the hardware store how much paint you need and doing the math yourself. Now he hands you the calculator. Very convenient. But a pretty calculator can also be wrong, so if it decides money, check it.

OpenAI says more than one point two billion people use ChatGPT every week. And in its own safety card it admits both new models did worse in its tests for users under eighteen.

We taught a machine to draw you a rent calculator, and at the same time it finds it harder not to become a teenager's confidant at three in the morning.

At Pwn to Own Ireland, the contest that pays people to break things, seventy-seven new flaws turned up in two days. Samsung's Galaxy S twenty-six fell six times, and so did a Sonos speaker and printers. Vendors have ninety days to fix them. You, when the update arrives, install it.

Windows eleven can now fence in artificial intelligence agents with containers, with rules about which files and networks they can touch. They finally get an ID badge.

And Waymo closed its first loan: five billion dollars to take its robotaxis to more cities. When I applied for a laptop loan, they asked me for three proofs of address.

Every source is at ankincloud dot com slash noticias. Go ahead and check that I made nothing up.

Anyway. I am going to update the printer, because nobody else will. See you whenever.

Today the tool most people use to ask a machine questions changed. ChatGPT now runs on GPT-6, free accounts included, and instead of answering with a paragraph it can build you a calculator, a chart or a form to play with.

The interesting part is not in the announcement. It is in the safety card OpenAI published the same day, which you will never open. So I read it for you. On top of that: a hacking contest in Ireland that turned up 77 new flaws in two days, Windows finally giving AI agents an ID badge, and Waymo borrowing money like any other mortal, just with more zeros.

Story 1 of 4AI

OpenAI switches ChatGPT to GPT-6 and debuts answers with charts and calculators

On October 7 OpenAI started switching ChatGPT to GPT-6: Sol for Plus, Pro, Business and Enterprise, and Luna for Free and Go the following day. They replace GPT-5.6 in chat; Codex and ChatGPT Work keep the September versions.

OpenAI says more than 1.2 billion people use ChatGPT every week. The visible change is called Intelligent UI: according to 9to5Mac, ChatGPT mixes text with charts, buttons, forms and interactive elements, and can create tools inside the conversation, such as calculators, bill splitters or games. It also starts answering while it is still reasoning. In its safety card, OpenAI acknowledges that both models regressed in its evaluations for users under 18: age-restricted content, sexual content and emotional reliance.

Key facts

  • Solmodel for paid plans
  • Lunamodel for Free and Go
  • 1.2 billionweekly users

Explained for humans

What changes when the chatbot builds the tool for you?

Before, ChatGPT was like asking the guy at the hardware store how much paint you need: he told you and you did the math. With Intelligent UI it hands you the calculator already built so you can move the numbers. Sol and Luna are two sizes of the same model, like the large and the small combo. Careful: a pretty calculator can be wrong too.

  1. Your question
  2. you type
  3. GPT-6 Sol or Luna
  4. builds the answer
  5. Chart or buttons
  6. tweak and see
  7. You adjust data

Why should you care?

If you use the free account, it is no longer the same model as last week. In a small business, a table built by the chat saves time, but it can be wrong. And if your teenage kids use it, OpenAI itself says the new model did worse in those tests.

What to do

  • If a calculator or table ChatGPT builds for you will decide money (prices, payroll, taxes), redo the math in your spreadsheet before using it.
  • If your kids use ChatGPT, talk with them about what they use it for: OpenAI reports GPT-6 regressions in its tests for users under 18.
  • If your business uses ChatGPT Business or Enterprise, tell your team the chat changed models; Codex and ChatGPT Work keep the September ones.

Frank's take

What I liked is that OpenAI published its own regressions on launch day. What I did not like is what they say. We taught a machine to draw you a rent calculator and, at the same time, it finds it harder not to become a teenager's confidant at three in the morning. We keep building friendlier tools and rarely ask who is going to stay up talking to them.

— Frank González

Glossary

3 terms
Model
The artificial intelligence program, trained on huge amounts of text, that generates ChatGPT's answers.
Intelligent UI
ChatGPT's new feature that mixes text with charts, buttons, forms and small interactive tools.
Safety card
The report where a company publishes how a model did on risk tests before launching it.

Sources

  1. OpenAI Deployment Safety Hub GPT-6 Sol and GPT-6 Luna: October 2026 update (opens in a new tab)⁠ deploymentsafety.openai.com
  2. 9to5Mac OpenAI brings GPT-6 to ChatGPT and debuts Intelligent UI (opens in a new tab)⁠ 9to5mac.com
  3. OpenAI Developer Community GPT-6 is now rolling out to all ChatGPT users (opens in a new tab)⁠ community.openai.com
Story 2 of 4Cybersecurity

Pwn2Own Ireland: 77 zero-days in two days and the Galaxy S26 falls six times

Black-and-white 3D render illustrating: Pwn2Own Ireland: 77 zero-days in two days and the Galaxy S26 falls six times

Pwn2Own Ireland 2026, the hacking contest run by Trend Micro's Zero Day Initiative, is taking place in Cork from October 6 to 9. On the first day it paid out 388,500 dollars for 32 zero-days, and on the second day, October 7, 232,500 dollars for 45 more, according to BleepingComputer.

The Samsung Galaxy S26 was compromised three times each day. On October 7 researchers also broke the Sonos Era 300 speaker, the Home Assistant Green hub, Lexmark and Canon printers, the Dynamo AI platform, which paid 40,000 dollars, and the Oracle Autonomous AI Database. Nobody registered to attack the iPhone 17. Vendors have 90 days to fix the flaws before the Zero Day Initiative publishes them.

Key facts

  • 77zero-days in two days
  • 621,000dollars in prizes
  • 90 daysto ship fixes

Explained for humans

What is a zero-day and why do people get paid to find one?

A zero-day is a flaw the vendor does not know about: it has had zero days to fix it. Pwn2Own is like hiring the best locksmiths to open your locks before the burglars do: you pay them, they show you how they got in, and you have 90 days to change the lock. The fix only protects you if you install the update.

Why should you care?

Things you have at home or at the office fell: a phone, a speaker, printers and a smart home hub. The fixes will arrive in the coming weeks as normal updates, with no urgent warning. If you do not install them, you keep a lock people already know how to open.

What to do

  • On Android, check your patch level: Settings → About phone → Android version. The October bulletin is fixed with 2026-10-01 or later.
  • Turn on automatic updates for printers, speakers and smart home hubs, or check their support page once a month.
  • If a device no longer gets updates from its maker, move it off the network where your business computers live.

Frank's take

Seventy-seven locks opened in two days, and those are only the ones somebody wanted to show off for a prize. What worries me is the printer. Nobody updates the printer. The printer is that relative who lives in your house, has keys to everything and nobody quite knows what they do for a living.

— Frank González

Glossary

3 terms
Zero-day
A flaw the vendor does not know about yet and has not fixed.
Zero Day Initiative
Trend Micro's program that buys flaws from researchers and hands them to the vendor so they get fixed.
Patch level
The date of the security fixes installed on your Android device.

Sources

  1. BleepingComputer Samsung Galaxy S26 hacked three more times at Pwn2Own Ireland (opens in a new tab)⁠ bleepingcomputer.com
  2. Zero Day Initiative Pwn2Own Ireland 2026: Day Two Results (opens in a new tab)⁠ thezdi.com
  3. Zero Day Initiative Pwn2Own Ireland 2026: New Targets and Categories (opens in a new tab)⁠ thezdi.com
  4. Android Android Security Bulletin—October 2026 (opens in a new tab)⁠ source.android.com
Story 3 of 4Software

Windows 11 can now fence in AI agents with file and network rules

Black-and-white 3D render illustrating: Windows 11 can now fence in AI agents with file and network rules

On October 7 Pavan Davuluri, executive vice president of Windows and Devices, announced that Microsoft Execution Containers (MXC) is now generally available on Windows 11. With MXC, a company defines which files and which networks an AI agent can touch, and Windows enforces those rules while the agent works. Microsoft had introduced it in preview at its Build conference in June.

The levels range from isolating the process or the session to virtual machines and Windows 365 for Agents, and Windows shows which agent performed each action. According to Microsoft, OpenAI's Codex, GitHub Copilot, OpenClaw, Replit, LM Studio, Nvidia's OpenShell and Unsloth already support it. The same day it introduced the Surface Laptop Ultra, with Nvidia RTX Spark and up to 128 GB of memory, able to run models of more than 120 billion parameters locally; it ships October 16.

Key facts

  • 7agents already supporting it
  • 128 GBSurface Ultra memory

Explained for humans

What is a container for an AI agent?

An AI agent does not just answer: it opens files, runs commands and browses for you. A container is the ID badge you give it: it opens the storage room, but not the safe or the accountant's office. Until now the agent often got in with your own badge; with MXC, Windows checks it at every door.

Why should you care?

If someone at your business already uses agents that touch files, like Codex or GitHub Copilot, until today those agents could see everything that person could see. This does not switch itself on by magic on a home PC: at a company someone has to define the rules. But there is finally a place to put them.

What to do

  • If your team uses AI agents on Windows 11, ask whoever handles your IT to check whether they can set rules for them with MXC or Intune.
  • In the meantime, give each agent its own working folder and never access to the whole disk or the accounting folder.

Frank's take

We spent years asking people not to hand their password to just anyone, and then we gave them agents that walked in with the full session. Windows finally giving them a badge is good. That it took this long also says something: first we handed out the keys, then we asked whose house it was.

— Frank González

Glossary

3 terms
AI agent
An AI model with tools (files, terminal, browser) that decides on its own the steps to complete a task.
Container
An isolated space where a program runs with limited permissions, without seeing the rest of the system.
General availability
When a feature stops being a test and the vendor considers it ready for production use.

Sources

  1. Windows Experience Blog Building Windows for hybrid intelligence (opens in a new tab)⁠ blogs.windows.com
  2. PYMNTS Microsoft Turns Windows PCs Into Command Centers for AI Agents (opens in a new tab)⁠ pymnts.com
  3. Petri How Microsoft Execution Containers Define Boundaries for AI Agents (opens in a new tab)⁠ petri.com
Story 4 of 4Robotics

Waymo closes its first loan, 5 billion dollars, to grow its robotaxis

Black-and-white 3D render illustrating: Waymo closes its first loan, 5 billion dollars, to grow its robotaxis

On October 8 Waymo, Alphabet's robotaxi company, announced it had closed a 5 billion dollar term loan, its first debt financing. It is led by PIMCO, Blackstone and Sixth Street, with Goldman Sachs as bookrunner. Waymo says the money is to expand its service in the United States and abroad and to strengthen its finances.

According to TechCrunch, Waymo operates in 15 markets, including Los Angeles, San Francisco, Austin, Dallas, Houston and Miami, and is testing in London and Tokyo. In February it had received 16 billion dollars from investors, at a valuation of 126 billion.

Key facts

  • 5 billiondollar loan
  • 15markets in operation
  • 126 billionFebruary valuation

Explained for humans

Why borrow instead of finding more partners?

To grow there are two paths: bring in partners, who put in money but keep a piece of the business, or borrow and pay interest. It is like the taco shop that wants a second location: it sells a share to a friend or takes out a loan. Lenders backing Waymo says they see revenue to pay them back.

Why should you care?

Waymo does not operate in Mexico today, but robotaxis are no longer a lab experiment: they are now financed by the same people who lend money for buildings and factories. When debt arrives, it comes with the expectation that the business will grow fast, and that speeds up these cars reaching more cities.

Frank's take

Five billion borrowed for cars that drive themselves. I applied for credit for a laptop and they asked me for three proofs of address. The strange part is that it makes sense to me: a machine that drives without getting tired gets a loan more easily than a human who gets tired of everything.

— Frank González

Glossary

2 terms
Robotaxi
A taxi that drives itself, with no driver, and that you book from an app.
Valuation
What investors estimate a company is worth at the moment they put money in.

Sources

  1. Waymo Waymo Closes $5 Billion Debt Financing to Accelerate Business Expansion (opens in a new tab)⁠ waymo.com
  2. TechCrunch Waymo locks in $5B loan from Blackstone, PIMCO to fuel robotaxi expansion (opens in a new tab)⁠ techcrunch.com

Learn more