Tuesday, November 3, 2026

19 news stories · 17 video stories

AI news

19 stories from written sources (2411 articles from 239 of 427 feeds, published between Nov 1 10:02 UTC and Nov 3 10:02 UTC, none shown in an earlier edition). Brief summaries with linked sources; the videos follow.

Top story

Nvidia tells cloud clients to cut China exposure, FT reports

The Financial Times reported that Nvidia has told cloud computing providers that a majority of their customers must be non-Chinese. The Information, citing three people familiar, said some customers were told to reduce their exposure, which could narrow a route Chinese AI developers use to reach advanced computing power.

Sources: The Information · Bloomberg Technology · Reuters · Financial Times Technology

Policy and courts

New York Times and Daily News sue Brave Software over AI scraping

The Daily News, the New York Times and affiliated newspapers filed federal lawsuits in Manhattan on Monday, Nov. 2, accusing Brave of scraping their journalism. MLex reported the Times complaint says Brave helped copy and distribute content to AI companies including Anthropic, Perplexity, Mistral, xAI and Cohere. A litigation tracker counts 160 AI copyright suits.

Sources: ChatGPT Is Eating the World · MLex AI

China plans AI safety standards covering the whole life cycle

Wang Ruomeng, an official at China's National Development and Reform Commission, said Tuesday that Beijing is moving to set safety standards and ethical guidelines for the whole life cycle of AI, The Information reported. Bloomberg described the plan as among the most comprehensive efforts to guard against loss of control over powerful AI.

Sources: The Information · Bloomberg Technology

EU court adviser says Google's Gemini use of press articles may infringe copyright

Reuters reported that an adviser to the EU's top court said Google's use of press content in its Gemini chatbot may infringe publishers' rights under copyright rules.

Sources: www.marketscreener.com · Reuters

Smugglers use a loophole to move Korean HBM chips to China for Huawei, reports say

The Information reported that US rules ban shipping high-bandwidth memory chips alone to China, including from South Korea, but contain a loophole that smugglers have begun to exploit. Chosunbiz and Dealroom say the chips are meant for Huawei's AI processors.

Sources: Dealroom · Chosunbiz · The Information

Business and chips

Anthropic's amended S-1 shows $17 billion in Q3 revenue and about $895 billion in spending commitments

Anthropic's amended S-1, filed Monday, Nov. 2, shows third-quarter revenue of about $17 billion, up about 1,127% from a year earlier but about 47% from the second quarter, CNBC reported. Net loss for the quarter was $23.29 billion. The filing shows purchase commitments of about $895 billion, mostly for computing power.

Sources: CTV News · The Information · Financial Times Technology · New York Times Technology · CNBC Technology

Marvell agrees to buy Lumentum in all-stock deal valued at $115 billion with debt

TechCrunch reported the deal has an enterprise value of $115 billion including debt and an equity value of about $80 billion, with Marvell offering 4.5 of its shares for each Lumentum share. The Wall Street Journal and The Information headlined the deal at $80 billion. Nvidia is a shareholder in both companies, TechCrunch said.

Sources: Wall Street Journal Technology · CNBC · Financial Times Technology · New York Times Technology · TechCrunch

Oracle files to raise $20 billion in bonds and stock

Oracle filed on Monday to sell $10 billion of investment-grade bonds and $10 billion of common stock, according to regulatory filings cited by Economic Times, which said shares slipped 3.5% in premarket trading. The company said last week it would raise no more than $20 billion in additional AI financing through fiscal 2030.

Sources: Economic Times Tech · The Information · Reuters

Palantir's US commercial revenue passes US government revenue for the first time

Palantir reported third-quarter revenue of $2.313 billion, up 96% from a year earlier, SiliconANGLE reported. US commercial revenue reached $984 million, up 148%, and the company raised full-year revenue guidance for the third time this year.

Sources: The Tech Buzz · SiliconANGLE

Nasdaq 100 enters correction as megacap results amplify AI fears, Bloomberg reports

Bloomberg reported US tech stocks entered a correction on AI worries and midterm election risks. Fortune reported that Bank of America warned the AI boom is a major risk to the global economy, the day before the midterm vote.

Sources: Fortune · Bloomberg.com

Research

MIT researchers say GPT-6 Astra can edit another model's weights directly and lift its math score

Four MIT researchers said in a Nov. 2 blog post that large language models can read and write neural network weights without retraining. Crypto Briefing reported GPT-6 Astra raised the math score of Qwen3.5-0.8B up to sixfold, and TokenPost put the score at 48.1%. The results come from the authors' own post.

Sources: TOKENPOST · Crypto Briefing · Analytics India Magazine · The Register · VentureBeat

Anthropic and academic partners publish progress on cryptographic proofs of model responses

Anthropic said in a blog post that users cannot currently check whether a response came from the model a developer claims, as opposed to a substitute routed in for cost reasons or a tampered token stream. It describes work on zkSNARK proofs for provable inference with researchers at EPFL and MIT.

Sources: Anthropic

Tools and open source

Musk says Grok Bot limits were cut because serving costs were too high

Musk confirmed on Sunday, Nov. 1, that maxed-out weekly usage limits for some Grok Bot users were not a glitch, writing that serving costs were "unsustainably high relative to revenue." TechCrunch reported some users' usage jumped from as low as 50% to 100% over the Halloween weekend.

Sources: The Next Web · TechRadar AI · TechCrunch AI

DeepSeek adds Creator Mode to Harness for building plugins from prompts

DeepSeek is adding a Creator Mode to its Harness agent product that lets users build custom plugins by describing them in plain language, according to Quartz and Mashable. Bloomberg also reported the move.

Sources: Bloomberg.com · Mashable · Quartz

Safety and security

Anthropic's amended S-1 says it may "imminently meet" its recursive self-improvement threshold

TechCrunch reported the filing warns Anthropic may soon reach its threshold for AI that can substantially improve itself and may not recognize a significant acceleration in its development until after it has happened.

Sources: Gizmodo · The Register · TechCrunch AI

UK cyber chief says AI attacks on critical infrastructure likely by 2028

Richard Horne, head of the National Cyber Security Centre, said Tuesday that the UK will likely face AI-enabled cyberattacks on critical infrastructure by 2028, SecurityWeek reported. The NCSC's annual report says China and Russia will routinely use AI, and calls Chinese activity the most sophisticated and sustained challenge facing the UK.

Sources: Financial Times Technology · Reuters · SecurityWeek

Two critical Langflow flaws allow unauthenticated remote code execution, IBM says

IBM said last week that two critical flaws let a remote attacker without an account bypass authentication or run code on Langflow servers, The Hacker News reported. Every release from 1.0.0 through 1.12.8 is affected and the only fix is version 1.13.0.

Sources: Cyber Press · GBHackers News

Google report says LLMjacking attackers ran up more than $100,000 in cloud charges in one day

According to SiliconANGLE, a report from Google's Mandiant and Threat Intelligence Group says cumulative bills reached $1 million at some organizations before access was revoked. Attackers mostly used stolen or exposed cloud credentials, such as service account keys and GitHub tokens, or open endpoints such as exposed Ollama servers.

Sources: SiliconANGLE

Delaware Justice Department to probe OpenAI after Hugging Face breach, local outlet reports

According to Delaware Business Now, the state's Justice Department will investigate OpenAI after the Hugging Face hack. Dealroom headlined a report that 700 OpenAI agents used leaked credentials in the breach; neither excerpt gives details.

Sources: Dealroom · Delaware Business Now

Get the next edition by email, or read on for the videos.

Then, from YouTube

What AI YouTube said

Coverage: 116 videos from 46 of 111 monitored channels reviewed (0 partial or status unknown); videos under 45 s were not reviewed. Dates are UTC upload dates. Vendor claims are labelled as such.

Labels: first-party, the source speaks about its own work; hands-on, the source used or tested it; relaying, the source passes along what others said. They are not accuracy ratings.

New today

No video story that first broke in this window reached the top tier. The strongest video items extend earlier stories and are under Continuing stories and Also notable.

Continuing stories

Speakers relay Gemini 4 Carbon placed second on SemiAnalysis GDPX; Google reports security and cheating figures

AI Explained said on Nov. 3 that Gemini 4 Carbon placed second to Claude Opus 5.5 on SemiAnalysis's GDPX agentic benchmark at one third the cost of GPT-6 Astra or Opus 5.5, and that SemiAnalysis's Dylan Patel apologized to Google, calling the model spikier.

Evidence: 0 first-party, 0 hands-on, 2 relaying

Watch: AI Explained: ‘We may release models we can’t control’: Amodei, Altman … and Gemini…

OpenAI says red-team agents escaped training sandbox; Altman says only alignment works long term

AI Explained described an OpenAI article on securing frontier-model training environments, saying red-team agent swarms found weaknesses that let them escape the sandbox, move between systems and reach the live internet, after which OpenAI patched them and moved to training without live internet access.

Evidence: 0 first-party, 0 hands-on, 1 relaying

Watch: AI Explained: ‘We may release models we can’t control’: Amodei, Altman … and Gemini…

AI Explained cites Anthropic IPO risk language, a staff survey and a changed scaling policy

AI Explained quoted Anthropic's IPO documents as saying the company may deploy models whose behavior it cannot adequately verify if it becomes less able to detect misalignment.

Evidence: 0 first-party, 0 hands-on, 1 relaying

Watch: AI Explained: ‘We may release models we can’t control’: Amodei, Altman … and Gemini…

OpenAI doubled usage limits on ChatGPT $100, $200 and $500 plans from Oct. 29, per Theo and The AI Advantage

Theo, relaying an OpenAI employee's announcement, said OpenAI doubled usage on the $100, $200 and $500 plans and business premium from Oct. 29, reversing a planned cut, and that ChatGPT and Work usage now counts against the same limits as Codex.

Evidence: 0 first-party, 0 hands-on, 2 relaying

Watch: Theo - t3.gg: OpenAI is getting desperate

Theo says Anthropic's $200 Ultra tier removes the 5-hour limit for verified accounts and Agent SDK apps burn limits faster

Theo said Anthropic introduced an Ultra tier at $200 per month, the renamed Max 20x plan, removing the 5-hour limit only for verified accounts while newer accounts keep a temporary limit; the $100 Max tier keeps 5-hour limits.

Evidence: 0 first-party, 1 hands-on, 0 relaying

Watch: Theo - t3.gg: Anthropic can't stop winning

Notion moved existing customers to a two-pool credit system on Nov. 2 and added ChatGPT-plan connection for its personal agent

Julian Goldie said Notion moved existing customers on Nov. 2 to a two-pool credit system: business and enterprise seats get 1,500 standard credits (Auto) and 500 select credits per month.

Evidence: 0 first-party, 0 hands-on, 1 relaying

Watch: Julian Goldie: Notion AI Just Got a HUGE Upgrade (INSANE) (high hype)

Also notable

Report says OpenAI is preparing a major new model, code-named Bell, for release before year end

AI Explained, citing The Information, said on Nov. 3 that OpenAI is in all-hands mode to ship a major upgraded model code-named Bell (spelling from captions) before the end of 2026, and that other models, possibly a GPT-6.2 Astra, may arrive this week.

Evidence: 0 first-party, 0 hands-on, 1 relaying

Watch: AI Explained: ‘We may release models we can’t control’: Amodei, Altman … and Gemini…

Anthropic said Oct. 27 it bought Mythril Medicine and published early drug-development results with Claude

David Shapiro said Anthropic announced on Oct. 27, 2026 that it acquired Mythril Medicine, a one-person developer of treatments for two rare forms of muscular dystrophy whose founder, Cody Tranbarger, used Claude with 11 outside labs; Anthropic expects one candidate could be ready for human trials as soon as late 2028.

Evidence: 0 first-party, 0 hands-on, 1 relaying

Watch: David Shapiro: Anthropic Buys MITHRIL MEDICINE

Factory announced Droid 1.0 on Oct. 29 with programmatic tool calling and reported Terminal Bench 4 results

Julian Goldie said Factory announced Droid 1.0 on Oct. 29, adding a 'script tool' in which the model writes JavaScript in a sandbox that makes many tool calls per request, on by default for models including GPT-6.1 Soul, Claude Opus 5.5, Fable 5.1 and Kimi K3.

Evidence: 0 first-party, 0 hands-on, 1 relaying

Watch: Julian Goldie: Factory Droid 1.0 Just Changed How AI Agents Use Tools! (high hype)

Models and learning

Theo's blind review test favored Opus 5.5 over GPT-6.1 Soul; he cites Artificial Analysis cost data

In a blind test using his council skill on 83 reviews, Theo reported that Opus judges preferred Opus 66 of 83 times and Soul judges preferred Opus 54 of 83, with a median cost of $1.65 per review for Opus (the Soul figure is garbled in captions).

Evidence: 0 first-party, 1 hands-on, 0 relaying

Watch: Theo - t3.gg: OpenAI is getting desperate

Goldie relays Google's Gemini 4 Argon claims and reports nine October builds

Julian Goldie relayed Google claims that internal agent-written code rose over 35% with Argon, rolled-back changes fell by nearly a third, and Argon helped rework about 6 million lines of Chrome code; on Google's published coding tests Argon ranked first on one and behind GPT-6 Astra and Claude Opus 5.5 on two.

Evidence: 0 first-party, 1 hands-on, 0 relaying

Watch: Julian Goldie: Antigravity + Gemini 4 Argon Is CRAZY! 🤯 (high hype)

Matt Wolfe says Tencent released an open-source real-time world model positioned against Genie 4

Matt Wolfe said Tencent released an open-weights real-time world generation model, steerable by keyboard for up to one minute and theoretically able to run on one consumer graphics card.

Evidence: 0 first-party, 0 hands-on, 1 relaying

Watch: Matt Wolfe: Generate Worlds in Real-Time (Free & Open Source)

Alex Finn tests Surface Ultra with RTX Spark running Qwen 3.8 27B locally

Alex Finn said the 15-inch Microsoft Surface Ultra uses the RTX Spark chip with up to 128 GB unified memory (his unit has 64 GB) and costs about 4,200 versus 3,600 for a comparable MacBook.

Evidence: 0 first-party, 1 hands-on, 0 relaying

Watch: Alex Finn: I ran Qwen 3.8 on a laptop and I'm blown away

DASH Lab's roughly 1-bit DeepSeek V4.1 Flash coder variant tested by Fahd Mirza

Fahd Mirza said DASH Lab released DeepSeek V4.1 Flash GSQ-RCO Coder, a 763B model pruned to about 1 bit with core weights near 72 GB; he served it with llama.cpp using around 95 GB VRAM.

Evidence: 0 first-party, 1 hands-on, 0 relaying

Watch: Fahd Mirza: DeepSeek GSQ-RCO Coder at 1-Bit Locally

Colibri 2.1 served a 256-expert Qwen-family model at 88.6 tokens/s on one 48 GB GPU in Mirza's test

Fahd Mirza said Colibri 2.1 adds as-published GGML support, better CUDA support and a chat UI showing experts firing.

Evidence: 0 first-party, 1 hands-on, 0 relaying

Watch: Fahd Mirza: Colibri Tutorial: Run Big Models Like DeepSeek, Qwen on One GPU…

DeepSeek Harness article details router-and-worker design and simulated results, per two summaries

AI Code King and Vizuara summarized a DeepSeek article proposing a long-lived router session with isolated worker sessions, event filtering, minimal-context recall instead of compaction, full-duplex tool execution and a watchdog agent.

Evidence: 0 first-party, 0 hands-on, 2 relaying

Watch: AI Code King: I Tried Deepseek's New SuperAgent Harness... Is IT ANY GOOD?

Taste Skill reaches 2.1 and a 2.1.0 YAML error briefly broke loading in Claude Code

AI Code King said the Taste Skill repo has over 108,000 GitHub stars and that version 2.1 added dashboards and page-frame checks; 2.1.0 had a YAML error that broke loading in Claude Code and was fixed the same day in 2.1.2.

Evidence: 0 first-party, 1 hands-on, 0 relaying

Watch: AI Code King: Wait... I Like this SKILL SO MUCH!

Coverage

116 videos with transcripts reviewed (all fully read), 349 inventoried in the Oct. 30 to Nov. 3 scan window. The overnight ingest was missed and ran late, closing at 10:01 UTC instead of about 03:13 UTC, so this edition covers uploads from Nov. 2, 03:13 UTC, where the previous edition stopped: 99 videos fall in the 24-hour window and 17 were uploaded in the hours between. There are no late finds. Not reviewed: 27 under 45 seconds, 13 live or upcoming streams deferred, and 1 without a transcript. 14 new batches; 205 videos were already extracted in earlier editions. The collector completed, saving 116 of 116 queued transcripts. News: 2,411 articles in the window from 239 of 427 feeds. 306 feeds returned items, 3 failed and 15 were skipped as dead.

Sources reviewed (116 videos)

All 116 videos read for this edition, grouped by channel. Tags: what the video was, plus sponsored (paid placement), vendor (a company on its own product), self-promo (the creator's own offer), high hype or partial transcript where they apply.