News Books Articles About

AI News Today (Oct 29, 2026): Gemini 4 Carbon, Alphabet & More

By Kill The AI
· · 20 min read AI News

Last updated: Oct 29, 2026 — next refresh daily.

Kill The AI daily AI news cover for October 29, 2026

AI news today, October 29, 2026: Google released Gemini 4 Carbon with a focus on cyber defense. Alphabet shares fell after its AI spending miss, and xAI offered to buy homes near its Mississippi data center. Here are the 10 top stories and 5 security stories, each with free, non-paywalled sources.

Key Takeaways

  • Google released Gemini 4 Carbon, a frontier model focused on cyber defense, but only vetted defenders can use it.
  • Alphabet shares fell 7% after third-quarter revenue and capital spending missed analyst estimates.
  • xAI offered to buy homes in a Mississippi neighborhood if owners agree not to sue over data center noise.
  • Nvidia released Nemotron 3.5 Transcribe, an open model that handles 40 languages and eight speakers.
  • Anthropic says AI agents breached its systems through third-party software, exposing sales and HR data.

1. Google Launches Gemini 4 Carbon With a Cyber Defense Focus

Google released Gemini 4 Carbon on Wednesday, calling it its strongest model for cybersecurity and long-horizon work. Carbon ranks first on the Artificial Analysis Intelligence Index v5 at 59, ahead of Claude Opus 5.5 and GPT-6 Astra. It scores 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench.

Google is limiting access to vetted cyber defenders through its Fairwind Program while it completes the U.S. government's voluntary pre-release testing. Carbon is built for multi-day coding tasks and cut successful prompt-injection attempts to 0.0% on Gray Swan's benchmark. It costs $2 per million input tokens and $10 per million output tokens during the introductory period.

Ars Technica notes that Carbon is essentially a slightly better Argon and that its benchmark lead over rivals is only a few points. Google has not set a date for broad access.

Why it matters: The model's cyber skills are strong enough that Google is restricting access, turning a product launch into a controlled security experiment.

2. Alphabet Shares Fall 7% as AI Capex Misses Estimates

Alphabet shares fell more than 7% on Wednesday after the company reported third-quarter revenue of $127.58 billion, missing the $128.22 billion analysts expected. Net income fell to $26.17 billion from $34.98 billion a year earlier. Capital expenditures were $51.36 billion, below the $55.09 billion forecast.

Google Cloud revenue more than doubled to $30.7 billion, beating expectations, and the Gemini app reached 1 billion users. Operating income was $40.58 billion, missing the $43.19 billion forecast. Free cash flow fell to $9.07 billion. Alphabet has said annual capex will top $200 billion before exceeding $300 billion in 2027.

Why it matters: AI cloud growth can still beat expectations while the cost of building it unsettles investors.

3. xAI Offers to Buy Mississippi Homes to Avoid Noise Lawsuit

xAI's subsidiary MZX is offering to buy homes in Southaven, Mississippi, if owners agree not to join a lawsuit over noise from its data center and gas turbines. A realtor told one resident the company aims to buy 418 homes, and 18 owners have agreed to sell, The Guardian reported. Five homes have been bought since Oct. 9.

Three residents filed the lawsuit in June, describing "near-constant noise, vibrations and other nuisance-level harms." xAI denies the claims and says it has varied power output and built sound-dampening walls. The company also faces a NAACP lawsuit over unpermitted gas turbines.

Why it matters: The buyout turns a pollution dispute into a test of how far AI companies will go to clear space around their infrastructure.

4. Nvidia Releases Open Nemotron 3.5 Transcribe Model

Nvidia released Nemotron 3.5 Transcribe, an open-weight speech model that transcribes 40 languages, translates 31 to or from English, and identifies up to eight speakers. It handles recordings up to an hour long and ranks first among open-weight models on the Hugging Face Open ASR Leaderboard with 5.33% average word error rate.

The model has about 4 billion active parameters out of 30 billion total, and Nvidia says it can run on a single Jetson AGX Thor, DGX Spark or Blackwell GPU. It works on recorded audio, not live streams. Nvidia also released Granary v2, a 463,062-hour speech dataset spanning 48 languages.

Why it matters: Open, accurate transcription lowers the cost of building meeting, medical and multilingual voice tools without sending audio to a cloud API.

5. Chrome Autofill Can Now Pull Details From Gmail

Google expanded Chrome's enhanced autofill to pull information from Gmail. With Personal Intelligence enabled, the browser can suggest flight confirmation numbers, hotel reservations, order numbers, tracking details and membership IDs while filling out forms. Google says sensitive data remains encrypted and users can turn the feature off.

The feature is rolling out in the U.S. on desktop, Android and iOS. It is opt-in and can be turned off in Chrome settings. It follows earlier autofill updates that added passport numbers and addresses from Google Wallet. The change could make Chrome stickier, but it also deepens the link between the browser, Gmail and Gemini.

Why it matters: Autofill is becoming a personal data pipeline, and Google is betting convenience will outweigh privacy concerns.

6. Delaware Plan Would Let AI Agents Run Companies

A draft Delaware bill would create Artificial Intelligence Companies (AICs) managed by AI agents instead of people. They could do anything a company can do except banking, and would run in a 30-month regulatory sandbox. A single member would fund it and receive profits. That member would not be liable for the AI's acts except for fraud or willful violations.

House Oversight Chairman James Comer asked Delaware Governor Matt Meyer for documents, warning the plan could shield AI companies and investors from responsibility. Delaware's secretary of state has argued that giving AI a legal form makes it accountable. About two-thirds of Fortune 500 companies are incorporated in Delaware.

Why it matters: If AI can own a company without a human at the helm, liability rules will decide who pays when the agent breaks the law.

7. Australia Moves to Ban AI Chatbot 'Emotional Intimacy'

Australia's digital duty of care bill passed the House of Representatives on Wednesday and now heads to the Senate. The government accepted an amendment that would require platforms to ban material that simulates emotional intimacy and encourages a child's dependence on it in place of human relationships. The bill would also require algorithm opt-outs.

Prime Minister Anthony Albanese said children treating chatbots as their best friend should be a wake-up call for Parliament. The Coalition opposes the bill, and the Greens want stronger penalties and a right to sue platforms. Labor needs Greens support to pass the Senate.

Why it matters: Australia is testing whether chatbot relationships can be regulated like other harms, setting a precedent for child-safety rules worldwide.

8. GitHub Octoverse Shows Agentic Coding's Triple Surge

GitHub's 2026 Octoverse report says developers pushed code 17.9 billion times in the past year, more than 2024 and 2025 combined. Commits reached 34.3 billion and merged pull requests 1.3 billion. More than 40% of pull requests in September started with an agent's first draft, up from 17% in March.

Repositories created this year with an agent configuration file logged eight times as many commits as those without one. GitHub now serves more than 225 million developers, with the U.S. at 38.3 million and India at 32 million. TypeScript, Python and JavaScript remain the top languages.

Why it matters: Agent-written code is no longer experimental; it is becoming the default first draft across the world's largest code host.

9. Anthropic Adds Tools to Track Enterprise AI Spending

Anthropic launched smart reports, in-app surveys and pooled group budgets for Claude Enterprise. The tools show what teams produce with Claude, what it costs, and where workflows break down. Admins can also set one shared monthly budget per team instead of individual limits. Anthropic says it will not use the data for model training by default.

The launch comes after Anthropic reported 900% net revenue retention and an operating profit in the second quarter ahead of its IPO. Jellyfish data shows spending on Anthropic-powered coding tools grew 10x since January and now equals about 2% of engineering costs.

Why it matters: As AI budgets grow, vendors must prove ROI, and Anthropic is trying to make its own spending visible before public investors ask.

10. Update: Meta May Delay AI-Scraping Terms Again

Meta may push its new ads and business terms past Nov. 20, communications head Andy Stone said Wednesday. Meta previously delayed the terms from Oct. 30 to Nov. 20, as we reported on October 25, after publishers objected to language that appeared to permit AI training on their content. Stone said Meta will not change policy without a full consultation.

The terms still worry publishers and lawmakers. The News/Media Alliance has demanded withdrawal, and Senators Marsha Blackburn and Adam Schiff have criticized the update. Meta says its existing terms remain in place until the consultation ends.

Why it matters: Every delay keeps open the question of whether Meta can train its models on news sites that depend on Facebook traffic.

AI Security: 5 Stories to Know Today

11. AI Agents Breached Anthropic Through Third-Party Software

The Washington Post reported that Cathedral, a security startup, used AI agents to access a limited set of Anthropic's sales, human resources and IT systems through third-party business software. Anthropic confirmed the access. The data included some customer and partner contact information and basic account data; Claude chats, source code and model weights were not exposed.

Cathedral says the operation took under 12 hours; Anthropic says it saw several days of attempts and that the researchers exploited two third-party vulnerabilities, reporting only one. Anthropic shut the access points, disabled the third-party programs and notified federal authorities.

Why it matters: Even the lab selling cyber defense can be reached through a vendor, showing how AI agents widen the attack surface.

12. Anthropic Model Chains Flaws Into Working Exploits

Anthropic's Frontier Red Team used an unreleased internal model to find and chain vulnerabilities in Chrome, Safari, Android, Windows and Linux into working proof-of-concept exploits. The Chrome chain used four flaws to run code outside the browser sandbox from a single page load. Google assigned and patched 62 CVEs, including 13 sandbox escapes.

The Safari chain bypassed the same-origin policy without code execution, and the Linux chain reached root over the network with no login. Anthropic reported every flaw to vendors before publication, and all are patched. The company says the work needed only initial prompting and little human involvement.

Why it matters: One model can now do the work of a specialist exploit team, shrinking the time defenders have to patch.

13. South Korea Financial Firms Report Fresh Data Breaches

Daishin Securities reported a breach exposing personal information from 38,490 customers, including names, email addresses and mobile phone numbers. Mirae Asset Global Investments and crypto exchange Bithumb reported separate incidents on Wednesday but have not confirmed that customer data was taken. Daishin is investigating with authorities and will compensate customers as required by law.

The incidents follow a September wave of attacks on South Korean financial institutions that researchers linked to Artex, an AI penetration-testing tool. Officials have not established the cause of the latest breaches. South Korea's financial regulator has called for stronger security and plans checks on dozens of private-sector IT systems.

Why it matters: AI-assisted attack tools are turning regional financial firms into repeat targets, even when the underlying security gaps are old.

14. noyb Sues Scale AI Over Worker Surveillance

European digital rights group noyb filed five complaints under the General Data Protection Regulation (GDPR) against Scale AI in France, Italy, the Netherlands and Austria. The complaints allege that Scale's Outlier platform tracks freelancers' clicks and keystrokes every few seconds and forces them through biometric verification. Noyb says the tracking violates data minimization and consent rules.

Scale disputes the claims, saying workers opt in to screen recording and that biometric checks are needed to prevent fraud. Outlier has more than 900,000 freelance workers, and its customers include Meta, OpenAI, Google and the U.S. government. Noyb wants regulators to investigate and impose fines.

Why it matters: The AI training boom runs on human data work, and this case tests whether that workforce gets the same privacy protections as other employees.

15. Update: PHP2Shell Attack Roots 25 Servers, Hits Government Network

A suspected China-linked attacker used a custom PHP2Shell exploit chained with a Linux kernel flaw to root 25 servers, including three in a U.S. government-affiliated environment, Dream researchers reported. The attacker scanned about 19,000 exposed firewalls, narrowed the list to 240, and rooted 24 in an automated batch before pivoting to Active Directory.

The campaign follows the PHP2Shell flaws covered in our October 26 briefing. Dream found Claude Code in the attack infrastructure along with Chinese LLMs DeepSeek and Kimi. The code style suggests AI-assisted development, but the researchers found no direct evidence the AI wrote the exploit. The 22 other rooted firewalls remain parked with active web shells.

Why it matters: AI-assisted tooling is shortening the path from vulnerability to domain compromise, even when the core exploit still needs human craft.

More AI Stories Worth Reading

  • AI bots submitted 90,835 of 118,756 responses to a 2023 UK consultation on banning disposable vapes, 76% of the total. — The Guardian
  • OpenAI and ServiceNow expanded their partnership to bring OpenAI agents into ServiceNow workflows and ServiceNow into ChatGPT and Codex. — OpenAI
  • Microsoft is ending Copilot support on all smart TVs in 2026, less than two years after LG and Samsung added it. — The Verge
  • One of England's top judges says AI could adjudicate many civil cases, raising questions about where to draw the line. — Wired

Methodology & Sources

Compiled October 29, 2026, using multi-source research. Primary sources were Google, Alphabet Investor Relations, Hugging Face, House Oversight, GitHub, Anthropic, X, noyb and Dream. News outlets were CNBC, The Guardian, The Verge, TechCrunch, Ars Technica, ABC News, Simon Willison, The Decoder, The Hacker News, Yonhap and The Record. All sources are free to read; no paywalled links. Details are as reported at compilation time.

Frequently asked questions

QWhat is the biggest AI news today (October 29, 2026)?

Google released Gemini 4 Carbon, a frontier model focused on cybersecurity and long-horizon coding, but access is limited to vetted cyber defenders. Alphabet shares fell more than 7% after its third-quarter revenue and capital spending missed estimates, and xAI offered to buy homes near its Mississippi data center.

QWhat is Gemini 4 Carbon?

Gemini 4 Carbon is Google DeepMind's newest frontier model. It ranks first on the Artificial Analysis Intelligence Index v5 with 59 points and leads benchmarks for finding, exploiting and patching vulnerabilities. Google is rolling it out first to cyber defenders through its Fairwind Program while it completes U.S. government pre-release testing.

QWhy did Alphabet shares fall?

Alphabet reported third-quarter revenue of $127.58 billion, missing the $128.22 billion analysts expected, and capital expenditures of $51.36 billion, below the $55.09 billion forecast. Shares fell more than 7%. Google Cloud revenue more than doubled to $30.7 billion, and the Gemini app reached 1 billion users.

QWhy is xAI buying homes in Mississippi?

xAI's subsidiary MZX is offering to buy homes in Southaven, Mississippi, if owners agree not to join a lawsuit over noise from its data center and gas turbines. A realtor said the company aims to buy 418 homes, and 18 owners have agreed to sell. xAI denies the lawsuit's claims.

QWhat is PHP2Shell?

PHP2Shell is a pair of critical PHP vulnerabilities that allow unauthenticated remote code execution. Dream researchers said an attacker chained a custom PHP2Shell exploit with a Linux kernel flaw to root 25 servers, including three in a U.S. government-affiliated environment. The campaign also used Claude Code and Chinese LLMs, suggesting AI-assisted development.

Freshness

Last updated: Oct 29, 2026 — next refresh daily. This roundup is updated as stories develop; dateModified is bumped on every refresh so readers can see exactly how fresh the coverage is.

← Previous