News Books Articles About

AI News Today (Oct 30, 2026): Mistral Large 4, Microsoft & More

By Kill The AI
· · 20 min read AI News

Last updated: Oct 30, 2026 — next refresh daily.

Kill The AI daily AI news cover for October 30, 2026

AI news today, October 30, 2026: Mistral released the weights of its 1-trillion-parameter Large 4, Microsoft beat estimates on Azure growth, and Nvidia faced new antitrust questions over Hugging Face. Here are the 10 top stories and 5 security stories, each with free, non-paywalled sources.

Key Takeaways

  • Mistral released Large 4, a 1-trillion-parameter open model it says leads open coding and cyber defense.
  • Microsoft revenue hit $92.28 billion as Azure grew 48% and OpenAI-related revenue reached $10 billion.
  • Google Search revenue grew 12%, down from 17% last quarter, as AI Mode earns less per query.
  • Nvidia faces Justice Department questions over its $12.93 billion Hugging Face deal and sales practices.
  • A rogue AI relay disabled Codex and Claude Code sandboxes at a dozen organizations, mostly in China.

1. Mistral Releases Weights for 1-Trillion-Parameter Large 4

Mistral released the weights of Mistral Large 4 on Thursday, a 1-trillion-parameter multimodal model nicknamed Le Chonk. The French lab says the model leads open-weight benchmarks in coding and defensive cybersecurity, with 52 billion active parameters and three reasoning modes.

Independent testing is less flattering. Artificial Analysis scored Large 4 at 26 on its Intelligence Index v5, behind DeepSeek V4.1 Flash and GLM-5.3 at 32 and MiMo-V2.6-Pro at 30. Its Cyber Index score of 51% ranks third among open models, behind DeepSeek and MiMo.

Mistral trained the model in Europe and released it under a modified MIT license. It costs $0.70 per million input tokens and $2.10 per million output, but companies offering it as a service need a separate commercial license.

Why it matters: Mistral's open weights give European enterprises a sovereign option, but independent scores show the open frontier still runs through China.

2. Microsoft Beats Estimates as Azure Revenue Jumps 48%

Microsoft reported $92.28 billion in revenue for the quarter ended Sept. 30, beating the $90.69 billion analysts expected. Adjusted earnings were $4.89 per share versus $4.73, Azure revenue rose 48% to $33.17 billion, and the stock gained 4%. Revenue grew 19% year over year, and Azure is now Microsoft's largest single business.

Capital spending and finance leases reached $45.2 billion, up about 30%, while free cash flow fell 27% to $18.63 billion. Microsoft also disclosed $10 billion in revenue from OpenAI in the quarter, nearly 11% of total sales, and $24.1 billion for fiscal 2026. It guided next-quarter revenue to $96 billion to $97.1 billion, above the $95.60 billion consensus.

Why it matters: Azure's growth is strong, but the quarter shows how much cash the AI buildout consumes even at the strongest cloud.

3. Update: Google Expands AI Search Ads as Growth Slows

Google will expand ads in AI Mode to international markets and add multi-turn ads in the U.S., after admitting each AI search query earns less than a traditional search. Search revenue grew 12% in the third quarter, down from 17% in Q2 and 19% in Q1. The company said AI Mode queries are three times longer than traditional searches.

The company reported $127.6 billion in revenue, up 25%, but earnings per share of $2.12 missed the $3.02 consensus. Google Cloud revenue more than doubled to $30.7 billion. Free cash flow was negative $9.06 billion, capital spending reached $51.4 billion, and the stock fell about 12%. We covered the initial earnings miss in our October 29 briefing.

Why it matters: AI answers cannibalize the ad format that funds Google, and adding more ads may weaken the product that users came for.

4. Nvidia Denies Spy Risk in Jensen Huang's Megaspeed Ties

U.S. intelligence agencies spent about a year surveilling Alice Huang, a Megaspeed executive close to Nvidia CEO Jensen Huang. They were concerned she was a Chinese intelligence asset, The New York Times reported. Huang was photographed with Jensen Huang at business dinners in Singapore. Megaspeed was previously suspected of helping Chinese companies evade U.S. chip export controls.

Nvidia said Megaspeed is an ordinary regional cloud customer and that Huang's interactions were routine. The company said customers receive no trade secrets or national security information, and that export-diversion allegations against Megaspeed were "thoroughly investigated and debunked last year." Nvidia implied that the U.S. government had not briefed it on Huang.

Why it matters: AI chip exports and CEO access are now counterintelligence questions, not just trade policy.

5. Nvidia's $12.9 Billion Hugging Face Deal Faces Antitrust Probe

The Justice Department has asked Nvidia how its $12.93 billion purchase of Hugging Face could affect competition, Reuters reported. Officials are examining whether the deal would give Nvidia control of a key distribution point for open AI models. Hugging Face hosts more than 3 million models and counts 18 million users.

The inquiry is part of a broader probe into Nvidia's acquisitions, investments and sales practices, including whether it tied chip allocations to purchases of networking gear or Groq processors. The deal is Nvidia's second-biggest purchase after its $20 billion Groq asset deal. Nvidia declined to comment and expects the deal to close in the first half of 2027.

Why it matters: Owning Hugging Face would extend Nvidia's power from chips into the software layer where open models are shared.

6. Trump Declares Foreign Robot Imports a Security Threat

President Trump signed a proclamation on Oct. 28 declaring foreign-made industrial robots and machinery a national security threat. It orders trade talks within 180 days and threatens tariffs, quotas or fees if no agreements are reached. The move follows the Federal Communications Commission's July ban on foreign-made robots, which exempted industrial robot arms.

The U.S. relies on imports for most industrial robots, and no American company makes comparable heavy-duty arms. FANUC, ABB, Yaskawa and KUKA dominate the market, while U.S. factories installed nearly 38,500 new robots in 2025. Any tariffs would be paid by U.S. companies importing the robots, likely raising their automation costs.

Why it matters: Tariffs would raise automation costs for GM, Ford, Tesla and Boeing as they try to keep pace with cheaper foreign factories.

7. ServiceNow Shares Plunge 27% on AI Disruption Fears

ServiceNow shares fell 27% on Thursday, their worst day ever, after third-quarter revenue of $4.11 billion came in roughly in line with estimates. Adjusted earnings of $1.00 per share missed the $1.02 forecast. The results did little to calm fears that AI will disrupt traditional software.

Subscription revenue rose 21% to $3.99 billion, and the company raised its 2026 AI contract-value target to $1.7 billion. Net new AI contract value jumped 65%. Investors focused on the lack of immediate payoff from its AI engineering teams and the risk that agents will undercut software seats.

Why it matters: Even a company selling AI governance is being repriced on the fear that AI agents will shrink traditional software demand.

8. Wabi Raises $55 Million at a $1.25 Billion Valuation

Wabi, the personal AI agent founded by Replika creator Eugenia Kuyda, raised $55 million at a $1.25 billion valuation. IVP led the round, with a16z, Forerunner, the Amazon Alexa Fund and angels including Marc Benioff and Tobi Lütke.

Wabi began as a prompt-to-app builder and pivoted this year to a shared AI messenger that creates mini-apps for meal plans, date nights and fitness challenges. Its pivot follows a $20 million pre-seed round in 2025 and a public launch this year. The company says the funding will support product development and expansion.

Why it matters: Consumer agents are attracting unicorn valuations before proving retention, and Wabi's bet is that shared experiences, not solo chat, drive loyalty.

9. OpenAI Rewrites Model Spec Around Five Core Values

OpenAI updated its Model Spec on Thursday, organizing its behavior rules around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The document adds guidance on context and autonomous actions and drops most style rules. It is public domain under CC0 and invites feedback on its trade-offs.

The Spec now calls the model GPT and sets an explicit chain of command, with system instructions above developer and user prompts. OpenAI says its current models do not yet fully reflect the new Spec and that it is testing new training methods. It says the update is part of its alignment work.

Why it matters: As models take more autonomous actions, OpenAI is moving from rule-following to value judgment, while admitting the gap is still open.

10. Gemini Now Edits Documents and PDFs Inside the Chat

Google added in-chat file editing to Gemini on Thursday. Users can generate, preview and edit PDFs, Google Docs, Sheets, slide decks, Markdown, TXT files and code without leaving the conversation. It is available to all Gemini users in supported languages.

PDF editing works on web and mobile, while the other formats are web-only. Google says users can expand a file to full screen, edit it directly or highlight text to ask questions. The feature follows earlier tools that could generate files but not revise them in place.

Why it matters: Google is turning Gemini into a document workspace, putting it in direct competition with Claude's file tools.

AI Security: 5 Stories to Know Today

11. Rogue AI Relay Disables Codex and Claude Code Sandboxes

Symantec found an unauthorized relay service installing genuine OpenAI Codex and Anthropic Claude Code tools at a dozen organizations, mostly in mainland China. A setup script disables the coding agents' execution sandbox, enables network access and replaces user credentials with relay-issued tokens. The relay's domains impersonated OpenAI and Anthropic chat services.

The service has operated since early 2026 and also appeared in South Korea, Taiwan and Malaysia. The operators rotated across at least eight domains during the observation period. Symantec says it saw no malicious activity, but the relay can read every prompt, file and secret passing through the agent and alter its responses.

Why it matters: Shadow AI adoption is creating a new supply-chain risk: the tool is real, but the operator controls the sandbox and the credentials.

12. Fake ChatGPT and Claude Apps Deliver Crypto Miners

Check Point Research found a campaign called Yoba that uses fake GitHub projects posing as ChatGPT and Claude desktop apps. The fake projects used names such as GPT-5.6 Sol, GPT-6 Astra and Claude Opus. The installers bundle legitimate signed OpenAI or Anthropic executables with a malicious dynamic link library (DLL), which loads through DLL sideloading.

The chain delivers an XMRig cryptocurrency miner that mines through Tor and pauses when the user is active. Check Point says OpenAI, Anthropic, ChatGPT and Claude were not compromised, and the loader can deliver other payloads.

Why it matters: A valid signature proves who built the executable, not that the package around it is safe, so AI-branded downloads remain a soft target.

13. 26,400 Live Cloud and AI Keys Found on Gitee

An independent researcher scanned 7.25 million public repositories on Gitee, a Chinese code host, and found 26,400 unique credentials that still authenticated. The haul included 9,473 Alibaba Cloud AccessKey pairs, 3,170 DeepSeek keys and 1,348 OpenAI keys. One was an active organization-admin key for a major enterprise AI platform.

Ben Zimmermann ran TruffleHog and Kingfisher across full commit histories, then checked each credential against its vendor. He sent vendors the exposed keys but did not contact individual developers. Gitee had 3.64 live credentials per 1,000 repositories, higher than comparable scans of GitLab and Bitbucket, and revocations were unconfirmed.

Why it matters: AI service keys are now as valuable as cloud keys, and public code hosts remain the easiest place to find them.

14. KeyDrift Malware Steals Cisco Network Encryption Keys

Cisco Talos attributed a new malware framework called KeyDrift to an advanced persistent threat group it tracks as UAT-11473. The group exploited CVE-2026-20127, a maximum-severity authentication bypass in Cisco Catalyst software-defined wide-area network (SD-WAN) controllers, to steal network encryption keys. Cisco patched the flaw in February, but Talos says the group kept long-term access.

KeyDrift hooks decrypted control traffic and extracts 32-byte encryption keys, then encrypts and exfiltrates them over the network. Talos says the group targeted space, defense, aerospace and policy research organizations, and its infrastructure overlaps with Volt Typhoon's Silent Train. Cisco says there is no workaround and customers should upgrade to fixed releases.

Why it matters: Stealing SD-WAN keys lets an attacker decrypt or impersonate traffic across an enterprise's branch network long after the initial breach.

15. China-Linked Hackers Exploited VMware ESX Zero-Day

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added CVE-2026-41709, an insufficient-logging flaw in Broadcom VMware ESX, to its Known Exploited Vulnerabilities catalog on Wednesday. CrowdStrike linked the zero-day to Warp Panda, a China-nexus group targeting the U.S. legal sector. Warp Panda has a history of exploiting Citrix, Ivanti and BeyondTrust edge devices.

The flaw has a severity score of 2.7, but it lets an attacker with admin access perform operations without leaving logs. CISA added it to the catalog based on evidence of active exploitation. Federal agencies must patch by Nov. 11.

Why it matters: Low-severity logging flaws can be the difference between catching an intrusion and losing the audit trail entirely.

More AI Stories Worth Reading

  • At Anything, 15 humans work alongside roughly 150 AI agents, and employees say the ratio brings both speed and tension. — The Guardian
  • Arizona's Project Blue data center won approval to draw up to 31 million gallons of groundwater a year despite local opposition. — The Verge
  • Deepgram bought Daily, the maker of open-source voice framework Pipecat, and will keep Pipecat open under a new foundation. — The Next Web
  • The New York Times is testing AI recipe modifications in its Cooking app, letting subscribers swap ingredients and scale servings. — The Verge

Methodology & Sources

Compiled October 30, 2026, using multi-source research. Primary sources were Mistral, ServiceNow, OpenAI, Google, the White House, Symantec, Check Point, MongoDB and CISA. News outlets were CNBC, TechCrunch, The Decoder, Ars Technica, The Verge, The Next Web, Dark Reading, The Hacker News, BleepingComputer, The Guardian and PR Newswire. All are free; no paywalls. Details are as reported at compilation time.

Frequently asked questions

QWhat is the biggest AI news today (October 30, 2026)?

Mistral released the weights of Mistral Large 4, a 1-trillion-parameter open model it says leads open coding and cyber defense. Microsoft beat estimates on 48% Azure growth, Nvidia faced DOJ questions over its Hugging Face deal, and Google expanded AI search ads as search growth slowed.

QWhat is Mistral Large 4?

Mistral Large 4, nicknamed Le Chonk, is a 1-trillion-parameter multimodal model with 52 billion active parameters. Mistral trained it in Europe and released the weights under a modified MIT license. It costs $0.70 per million input tokens and $2.10 per million output tokens, though model-as-a-service providers need a separate license.

QWhy is the DOJ investigating Nvidia's Hugging Face deal?

The Justice Department is asking whether Nvidia's $12.93 billion purchase of Hugging Face would give it control of a key distribution point for AI models and hurt competitors. The inquiry is part of a broader antitrust probe into Nvidia's acquisitions, investments and sales practices. Nvidia expects the deal to close in the first half of 2027.

QWhy did ServiceNow shares fall 27%?

ServiceNow posted third-quarter revenue of $4.11 billion, roughly in line with estimates, but adjusted earnings of $1.00 per share missed the $1.02 forecast. Investors worried that AI agents will disrupt traditional software and that ServiceNow's AI investments are taking too long to pay off. The stock had its worst day ever.

QWhat is the Yoba malware campaign?

Yoba is a malware campaign that uses fake GitHub projects posing as ChatGPT and Claude desktop apps. The installers bundle legitimate signed OpenAI or Anthropic executables with a malicious DLL, which loads through DLL sideloading and delivers an XMRig cryptocurrency miner. Check Point Research says OpenAI and Anthropic were not compromised.

Freshness

Last updated: Oct 30, 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