OpenAI now lets you text your ChatGPT Dot agent
The company is rolling out text messaging for its Dots, allowing US Pro subscribers to assign tasks, share photos or voice notes, and receive proactive updates through iMessage or RCS.
Alleged Chinese spy had ‘close relationship’ with Nvidia boss Jensen Huang 😱
The New York Times reports that US intelligence officials believe an executive at one of Nvidia’s data center customers, Megaspeed, may be a “Chinese intelligence asset,” and has a “close relationship” with Nvidia CEO Huang. Details are sparse in the article, especially on how close that relationship was, Nvidia says she’s just one of “thousands of customer executives” Huang meets every year.
Instagram head warns backlash against AI will continue to grow
Adam Mosseri says public frustration with AI is likely to intensify over the next few years, driven by concerns about job displacement, content ownership, safety and the spread of misleading AI generated material.
Chrome can now use Gmail to autofill online forms
Google is updating Chrome with AI powered autofill that can retrieve details from Gmail, including flight confirmation numbers, membership IDs, delivery tracking information, addresses and payment details, to help users complete online forms.
The feature is rolling out to US users on desktop and through Gemini in Chrome on iOS and Android, with an option to enable it in settings and Google saying sensitive information is encrypted.
Google Launches Gemini 4 Carbon
Google has introduced Gemini 4 Carbon, a new frontier model focused on coding, agentic tasks, and cybersecurity that the company says outperforms the top models from OpenAI and Anthropic on some benchmarks.
- Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index v5.0, ahead of Claude Opus 5.5 and GPT-6 Astra.
- Google says Carbon delivers significant improvements in coding and agentic tasks, as well as in key cybersecurity defense capabilities, particularly in identifying software vulnerabilities and generating more secure code.
- The model currently leads SEC-bench Pro, which evaluates AI models on finding real-world software bugs, and ExploitGym, which measures their ability to develop exploits from vulnerabilities.
- Google says Carbon can handle longer, more complex engineering tasks, with internal teams reporting that it can work for days while navigating setbacks and completing multiple steps.
- The company also reports improvements in resisting prompt injection attacks, avoiding shortcuts, and collaborating with users. Compared with Gemini 4 Argon, Carbon reportedly cuts unsupported claims by half.
Access is currently restricted to selected cybersecurity defense companies. Google plans to expand availability to additional partners through its Fairwind Program. It will launch at the same introductory API prices as Argon: $2 per million input tokens and $10 per million output tokens.
Amazon Introduces Five New Alexa Devices
Amazon has announced five new Alexa-powered devices as it expands its smart home lineup. The releases include a redesigned Echo Show 16, a new Echo Studio speaker, and devices designed for sleep tracking and displaying photos.
- The redesigned Echo Show 16 features a 16-inch screen, built-in Fire TV, and Amazon’s first Echo Show camera with night vision. It is available for preorder starting at $499.99.
- The new Echo Studio costs $249.99 and includes a camera that allows Alexa to recognize objects users hold up to it. It also lets users remotely check on a room.
- The latest Echo Dot costs $99.99 and adds a built-in smart home hub and Wi-Fi extender for the first time.
- Echo Time is a $199.99 bedside clock with sleep-sensing technology designed to wake users during lighter stages of sleep without requiring a wearable device.
- The Alexa Photo Frame starts at $199.99 and automatically syncs pictures from Amazon Photos.
The launch comes as Amazon faces growing competition in the smart home market. Apple recently introduced its $299 HomeView smart home device with a 7-inch screen and face recognition, alongside the $149 HomePod mini 2.
Turn Your Photos into Halloween Horror Scenes
Go to ChatGPT or Gemini (or any other AI of your choice) and try these prompts:
Make Your Own Scary Movie Poster
Create a photo of me in a dream Y2K style portrait of me laying in shiny purple bedding as I hold a large 90s-style landline in a daydreaming pose. I wear delicate jewelry and gold chunky rings. The room behind me is girly with 90s-style posters. My makeup is simple yet glamorous with pink lip gloss. The photo should have a grainy 90s style with a light source like a lamp in a dimly lit room at night. A 90s photo album rests beside me with a disposable camera. [Ghostface] stands behind me staring at me, and it should be standing in the doorway of a dimly lit hallway. The background behind it should be slightly dark and ominous. Vertical poster format, 2:3 aspect ratio.
You can replace [Ghostface] with any character or figure you want, as well as make any other changes you see fit.
Audiences Aren’t Buying the AI Movie
A $200 meme movie out-earned China’s $1.5 million AI epic by more than 50 times.
- Bona Film Group’s fully AI-generated “Sanxingdui: Future Past” pulled in just $181,000 despite 10,000-plus showings on opening day. Nearly 100 people spent two years making it.
- Earlier this year, a hand-drawn animated film called “Niu Lai” cost $200, went viral for its rough quality and cleared $10 million. Viewers paid for visible human effort and skipped the polished machine output.
- Studio exec Jiang Defu calls the flop the price of going first. One viewer told AFP a future version might earn a ticket.
Who buys a seat for a film that was built to skip the human part?
This week, put one obvious human fingerprint on your AI-assisted output, like a hand-made thumbnail or your own face on camera, and compare how it performs.
Microsoft Wants $25B to Fund Its AI Push
The richest cash machine in tech is going to the bond market to pay for AI.
- Microsoft is marketing at least $25 billion in bonds across as many as ten tranches, with maturities from two to 40 years. It’s the company’s first US investment-grade sale since 2017, when it raised $17 billion.
- Its latest quarter shows the pull: $45.2 billion in capex against $18.6 billion in free cash flow. Spending has outrun even Microsoft’s income, and the sale is meant to cover the gap for general corporate purposes.
- Final size isn’t set. The longest bonds, running to 2066, were pitched at up to 135 basis points over Treasuries. A AAA rating buys Microsoft that price, while CoreWeave’s yields have already pushed toward 15%.
That $313 billion in hyperscaler bonds is nearly three times the 2025 total.
This week, find the AI vendor you depend on with the weakest balance sheet and sketch a fallback before their funding costs land in your API bill.
Gemini 4 Carbon Is No. 1 and Private
The top-ranked AI model right now is open to a small group of security teams and nobody else.
- Google released Gemini 4 Carbon to select cyber defenders two weeks after Argon reached paying users. It scores 59 on the Artificial Analysis index, one point clear of Claude Opus 5.5.
- Google says Carbon hit 88.2% on SEC-bench Pro for finding real vulnerabilities and 71.0% on CWE-bench for patching them, with guardrails off. Its Chrome team rated Carbon’s patches more than 10 times better than another lab’s model, which Google didn’t name.
- Argon took 13 days to go from defenders to paying users. Google promises Carbon a wider release as soon as possible, at $2 and $10 per million tokens, with a footnote that lifts it to $4 and $20 later.
Bug hunters get the sharp tool first.
This week, patch the dependencies you’ve been ignoring, because AI vulnerability finders are about to get a lot better.
Google’s New AI Hunts Software Bugs
Google has unveiled Gemini 4 Carbon, its new frontier model built for long coding and cybersecurity work. It arrives just two weeks after Gemini 4 Argon.
- Carbon scored 88.2% on SEC-bench Pro, a test of finding real software vulnerabilities, and 71% on CWE-bench, which measures how well it writes fixes.
- It ranks #1 on the Artificial Analysis Intelligence Index with a score of 59.
- Coding also improved: Carbon jumped 89 points on WebDev Arena to reach #4.
- Access starts with selected cyber defenders, then more trusted partners through Google’s Fairwind program. Regular users and developers can’t use it yet.
- Introductory API pricing is reported at $2 per million input tokens and $10 per million output tokens.
A model this good at finding security holes can protect software or attack it.
That’s why Google is giving defenders a head start before anyone else gets access, a release pattern we’re now seeing across the biggest AI labs.
40% of Pull Requests Are Now Agent-Written. Who Is Actually Reviewing Them
GitHub's Octoverse report says AI agents originated more than 40% of all pull requests in September, up from 17% in March. That number will likely cross 50% before the end of the year. Meanwhile, Google's Gemini 4 Carbon — a model built to find real software vulnerabilities and produce working fixes — is now in the hands of cyber defenders. The combination creates a question every engineering team needs to answer honestly: if the code is increasingly written by agents, are the review processes, security gates, and testing frameworks built for that reality, or for the old one? Ask your AI 1. If 40% of our pull requests were written by AI agents right now, what would break in our current code review process? Walk me through every step from PR open to merge — and flag where a human reviewer would be unable to verify agent-authored code the same way they verify human-authored code. 2. Gemini 4 Carbon scores 71% on producing working fixes for real CVEs. If a model like this were applied to our codebase today, what are the top five categories of vulnerability it would most likely surface — based on our stack, our dependencies, and any CVEs disclosed in tools we use this week (Red Hat AI Gateway, MLflow, MCP Toolbox)? 3. Design a code review policy for agent-authored PRs that is different from our current human-authored review policy. What checks should be automated, what should require human sign-off, and what should be blocked entirely until a security review is completed?
Apple Puts AI Inside Pages, Keynote and Numbers
Apple is adding a built-in AI assistant to its iWork apps, so you can draft documents, design slides and analyze spreadsheets without leaving the app.
- The assistant can draft or restructure documents in Pages, design slides in Keynote, and suggest charts from your data in Numbers.
- It runs on Apple’s own models plus OpenAI, and launches as a beta.
- It’s part of Apple’s Creator Studio subscription at $12.99/month, rolling out in November.
- At launch it works in US English only, on devices that support Apple Intelligence.
- With macOS 27, the apps also support MCP, so outside AI agents like Claude or ChatGPT can connect to them.
Microsoft has Copilot in Office and Google has Gemini in Docs.
Now Apple’s office apps are catching up, and MCP support means you won’t be stuck with Apple’s assistant. You can plug in the AI you already use.
Claude Now Shows Bosses What AI Is Worth
Anthropic has launched new reporting tools for Claude that show companies what their AI spending actually produces.
- Managers can pick a date range and see what a team produced, what it cost, and where its workflows got stuck.
- The reports also suggest ways to improve results.
- Teams can share pooled token budgets to keep spending under control.
- Anthropic says this data won’t be used to train its models by default.
- A Jellyfish survey found Claude-powered tools are now behind nearly a third of code changes submitted for review, with spending up 10x this year.
Companies have been spending fast on AI without clear proof it pays off.
By showing cost next to results, Anthropic makes it easier for finance teams to justify, and keep paying for, Claude.
Check If AI Agents Can Use Your Website with Dedica
AI agents are starting to browse, book and buy for people. Dedica runs 125 checks on your site to find what those agents can’t do, then tells you how to fix it.
Steps to Follow:
- Open Dedica and paste your website URL.
- Run the free check.
- Review the tasks AI agents couldn’t complete.
- Apply the suggested fixes, starting with your signup or checkout page.
- Run the check again to confirm.
Apple launches touchscreen MacBook Pro
Apple unveiled a new MacBook Pro with an OLED touchscreen, the first Mac with touch. It starts at $2,899 (14-inch) or $3,399 (16-inch), and preorders are open ahead of a November 6 release.
Apple builds the touch sensor directly into its new tandem OLED panel to cut latency, and pitches uses like testing iPhone and iPad apps or playing instruments onscreen. Many existing apps work with touch immediately.
Apple spent over a decade calling this a bad idea. Steve Jobs called touchscreen laptops “ergonomically terrible” in 2010, and Apple killed its Touch Bar compromise in 2023 after seven years.
Dutch government plans 25,000 job cuts with AI
The Dutch government plans to cut 25,000 civil service jobs, more than 15% of its workforce, over several years as it looks to streamline operations and expand AI use.
Officials aim to save €900 million annually by 2027 and €3.5 billion by 2035. The plan involves wider AI adoption and less bureaucracy, while exempting police and defense from cuts.
Some departments could lose 45% to 60% of staff, while officials target a 20% productivity boost through AI. That’s a massive bet on technology keeping public services running smoothly.
Google’s Gemini 4 Carbon tops AI benchmarks
Google just announced Gemini 4 Carbon, a new frontier model built for cybersecurity and long coding tasks. It scored 59 on Artificial Analysis’ Intelligence Index, taking the top spot from Anthropic’s Claude Opus 5.5.
Carbon lands two weeks after Argon’s launch, and Google says it leads key cyber tests, including 88.2% on SEC-bench Pro. Its $2/$10 per-million-token intro price is half Opus 5.5’s, before doubling to match it.
So the world’s top-ranked model is currently off-limits to almost everyone. Only a small group of cyber defenders gets access this week, and Google hasn’t set a date for developers or consumers.
DoorDash to buy Instacart for $17 billion
DoorDash agreed to buy Instacart for about $17 billion, or $65 per share, a 26% premium to Instacart’s pre-rumor price. The deal, at least 75% cash, should close between late 2027 and mid-2028.
DoorDash wants to offer grocers Instacart’s store tech plus its own delivery network. Meanwhile Instacart’s share of delivery app spending slid from 26% to 18% since 2021, while DoorDash’s climbed to 59%, per Consumer Edge.
Instacart has come a long way down. The two companies discussed a deal back in 2021 that could have valued Instacart at $40 to $50 billion. This deal comes in at well under half that.
Amazon adds 100+ shopping features to Alexa
Amazon added more than 100 new capabilities to Alexa for Shopping ahead of the holidays. You can now manage lists and registries, change subscription timing, update delivery instructions, and sort out payment issues.
Amazon merged its Rufus chatbot and Alexa+ into Alexa for Shopping in May and keeps adding agentic tools, like auto-buying items at target prices. It says agentic feature use has grown 8x this year.
Amazon keeps raising the scoreboard. It claimed 350 million Alexa shoppers over 12 months on its Q2 earnings call, and now says over 400 million, just in time for holiday spending.