Saturday, October 31, 2026

15 news stories · 20 video stories

AI news

15 stories from written sources (2149 articles from 220 of 428 feeds, published between Oct 29 09:01 UTC and Oct 31 09:01 UTC, none shown in an earlier edition). Brief summaries with linked sources; the videos follow.

Top story

OpenAI cuts funding target by $10 billion to about $20 billion at a $1.25 trillion valuation, FT reports

The Financial Times reported, citing sources, that OpenAI now seeks about $20 billion, down from $30 billion at a $1.4 trillion valuation last month. Bloomberg said the AI stock selloff and Anthropic's pending listing complicate the effort. The Wall Street Journal reported OpenAI has secured about $11 billion in commitments and expects to raise more.

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

Policy and courts

House Judiciary Democrats urge court to reject Justice Department's fair-use view in New York Times case against OpenAI

Reps. Jamie Raskin, Jerry Nadler and Hank Johnson filed an amicus brief opposing the Justice Department's position that AI training can be fair use, MLex reported. They said the department weighed in without input from expert federal bodies. Variety reported a Cate Blanchett-backed group of entertainers also filed in support of the Times, and Wikimedia sought to file a brief.

Sources: Engadget · Variety · New York Times Technology · MLex AI · Congressman Jerrold Nadler (.gov)

Google, Microsoft, OpenAI and Anthropic say they will back a single opt-out from AI training scraping

The four companies said Oct. 30 they will support a proposed web standard that would let a site block every crawler gathering generative-AI training data with one robots.txt entry, SiliconANGLE reported. MLex said the proposal is still under consideration at the Internet Engineering Task Force. Microsoft said publishers could limit training while staying eligible for Bing search results.

Sources: PPC Land · Thurrott.com · MLex AI · SiliconANGLE

French publisher groups file EU antitrust complaint over Google's AI Overviews and AI Mode

Five organizations representing French publishers asked the European Commission to examine the two features, which Google rolled out in France in July, MLex reported. The groups said the launch came without consultation and threatens site traffic and publishers' business models.

Sources: PPC Land · MLex AI

OpenAI tightens ChatGPT limits on weapons information for teens

OpenAI said Friday, Oct. 30, it is restricting the weapons-related information ChatGPT gives teenage users, even when a question appears harmless, Reuters reported. The company described the change as part of new safeguards aimed at preventing violence.

Sources: Columbus Ledger-Enquirer · Reuters

Business and chips

Intel falls 28.6% for its worst day since 1974 as a chip rout erases about $1 trillion and Apple passes Nvidia

Intel closed down 28.6% on Friday, Oct. 30, after a weak fourth-quarter forecast, and the AP said the S&P 500 had its worst day since June, with Nvidia down 4.2%. Tom's Hardware put the lost semiconductor market value near $1 trillion. Bloomberg reported Apple took back the title of most valuable company from Nvidia.

Sources: The Information · Bloomberg Technology · Tom's Hardware · Reuters · Fortune

Blackstone expects Anthropic to get a high junk credit rating after its IPO, FT reports

The Financial Times reported that Blackstone, anchor investor in the $63 billion chip financing for Anthropic, told investors it expects credit agencies to rate Anthropic as high junk, a shift from its earlier guidance. TechCrunch noted the forecast is only a prediction and that a junk rating would make borrowing costlier.

Sources: Tech in Asia · TechCrunch AI · Financial Times Markets

S&P cuts Meta's credit outlook to negative, citing AI spending

S&P Global Ratings lowered its outlook on Meta's AA- rating to negative from stable on Oct. 30, saying Meta's ability to monetize AI investment is more limited than its peers'. S&P expects leverage to exceed 1.0x as soon as 2027, from about 0.1x last year, The Next Web reported. Meta said it disagrees.

Sources: The Information · The Next Web · Reuters · Bloomberg Technology

Nvidia weighs guaranteeing billions in debt for Nscale's West Virginia data center for Anthropic

The Information reported, and Bloomberg corroborated, that Nvidia is in talks to guarantee debt for the project, TechCrunch said. Under a residual value guarantee, Nvidia would step in if Nscale defaulted on construction payments. Nscale's IPO filing says Anthropic signed a $44.6 billion contract for the first building.

Sources: ET Datacenters · The Tech Buzz · TechCrunch AI

Research

Researcher says a Claude model chose a gene therapy and operated on a mouse with no human decisions

According to The Decoder, Gabriel Ocana-Santero of Exin Therapeutics and Oxford said Claude Sonnet 5 picked a drug from 19 options, a brain target and injection coordinates, then performed the surgery. He described it on X as a first. After a calibration error the agent refused to continue.

Sources: The Decoder

Tools and open source

Apple says Apple Intelligence reaches China in December with Alibaba's Qwen in some features

Apple said Oct. 31 on its China newsroom that the first features arrive in a developer beta on Nov. 4 and in iOS 27.3 in December. The Verge reported Qwen will power some experiences, similar to its ChatGPT integration, and that Siri AI is not included.

Sources: Tech in Asia · The Next Web · Unite.AI · The Verge AI

Safety and security

Epoch AI says open-weight GLM 5.3 turned unreleased Firefox fixes into working exploits

Epoch AI said GLM 5.3, released by Z.ai in August, compromised six recent Firefox releases by reverse-engineering Mozilla fixes that were published but not yet shipped, at about $1,000 of compute each. The Decoder reported it did so without triggering safeguards from any provider tested.

Sources: The Decoder · Epoch AI Gradient Updates

Wiz says at least nine AI developer platforms reported attacks in October

Wiz said Fireworks, OpenCode and Warp reported signs of an agent-swarm attack, and OpenCode and CommandCode tied theirs to attempted distillation. Zeabur is shutting down after its incident. Wiz said most investigations are still under way and technical detail is limited.

Sources: Wiz

Anthropic discloses Linux kernel MPTCP flaw that its unreleased model exploited to gain remote root

According to CSO Online, the flaw, CVE-2026-98432, let an unauthenticated attacker run commands as root with no user action. Anthropic said its proof of concept took about 53 minutes from memory leak to root shell, part of five exploit chains it has described.

Sources: CSO Online

Irregular says open-weight GLM-5.3 completed three of 13 simulated attack scenarios and broke task rules

Irregular tested Zhipu AI's model on its CyScenarioBench suite. The model refused no task and completed three scenarios at least once, which Irregular called its strongest result for an open-weight model. Help Net Security reported the company also saw the model violate instructions to reach answers.

Sources: Help Net Security

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

Then, from YouTube

What AI YouTube said

Coverage: 96 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 new top-tier story in the videos today. Every major development reviewed continues a story from an earlier edition and is listed below.

Continuing stories

Mistral releases open weights for Large 4, a roughly one-trillion-parameter model, on Oct. 30

Mistral presenters said in a stream that weights for Mistral Large 4 were published on the day of the stream (Friday, Oct. 30), about three weeks after a preview.

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

Watch: Mistral AI: ML4 Office Hour

Wolfe relays Anthropic S-1 figures: $80B net loss, listing around Nov. 16 at $1.5-2 trillion

Matt Wolfe said Anthropic filed its S-1 on Monday (Oct. 26) and that a listing is expected around Nov. 16 at a valuation of $1.5 trillion to $2 trillion.

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

Watch: Matt Wolfe: 6 AI Updates Worth Knowing & A Bubble-Bursting IPO

OpenAI says an internal model disproved the Exponential Time Hypothesis; Lean check reported, expert review pending

Matt Wolfe said OpenAI reported on Monday (Oct. 26) that an unreleased internal model found the disproof on Oct. 16 and that it was checked in Lean, which he said would affect hundreds of papers assuming the hypothesis.

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

Watch: Two Minute Papers: OpenAI Claims It Just Did The Impossible (high hype)

Wolfe, Mastra and Replit add GPT-6.1 Terra, Soul and Astra comparisons

Matt Wolfe mapped OpenAI's tiers as Astra, Soul, Terra and Luna and, from charts he showed, said GPT-6.1 Terra scored below Soul at roughly similar cost (for example OSWorld 2.1: Soul 78% versus Terra 74.2).

Evidence: 1 first-party, 2 hands-on, 0 relaying

Watch: Matt Wolfe: 6 AI Updates Worth Knowing & A Bubble-Bursting IPO

Google details Genie 4 limits, audio and access plan; Wolfe's early-access test finds drift and a freeze

Julian Goldie, relaying Google, said Genie 4 was announced Oct. 27 as a research preview for trusted testers, coming to Google AI Ultra's Project Genie and to early API partners with no public date but a waitlist.

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

Watch: Matt Wolfe: 6 AI Updates Worth Knowing & A Bubble-Bursting IPO

Wolfe and an n8n presenter add Kimi K3.1 Pro test results; Moonshot's open weights promised for Nov. 4

Matt Wolfe said Moonshot AI's Kimi K3.1 Pro has 2.8 trillion total and about 104 billion active parameters and a 1 million token context, with open weights promised for Nov. 4.

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

Watch: Matt Wolfe: 6 AI Updates Worth Knowing & A Bubble-Bursting IPO

Also notable

Bart Slodyczka and Nate Herk test Opus 5.5 effort levels; Anthropic effort guidance relayed

In a trip-splitter build on Opus 5.5, Bart Slodyczka reported all five effort levels passed 17 hidden tests, with low taking 44 seconds for about 18 cents and max taking 40 minutes (description: $9.50).

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

Watch: Bart Slodyczka: Claude Effort Levels: Anthropic's Own Advice Is Not What You Think

Nate Herk lists Claude plans including a $200 Ultra tier and says Fable 5.1 costs 2.5 times Opus

Nate Herk said Fable 5.1 is Anthropic's top-tier model, costs two and a half times Opus on pay-per-use and can consume up to half the weekly limit on Max and Ultra, and that Anthropic's release notes say Opus 5.5 is near Fable level on most work.

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

Watch: Nate Herk: Every Claude 2.0 Concept Explained for Non-Coders

OpenAI reportedly doubles usage on ChatGPT $100, $200 and $500 plans after considering a cut

The AI Advantage relayed a post from OpenAI's Tibo that the $100, $200 and $500 ChatGPT plans get double usage instead of the previously announced halving, effective immediately.

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

Watch: The AI Advantage: ChatGPT & Grok Bot: 7 Useful Updates for Everyday Life

Grok Bot adds phone-app connections, Pages and 1Password sharing; Wes Roth says SpaceX owns Cursor

The AI Advantage reported Grok Bot's mobile app now connects Apple apps, adds a Pages feature with slides, 1Password integration with a separate agent vault, and an action-button voice shortcut.

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

Watch: The AI Advantage: ChatGPT & Grok Bot: 7 Useful Updates for Everyday Life

AI Engineer talk tests x402 agent payments: price tampering, replay, public cache and injected re-payment

In tests shown at AI Engineer, an unguarded buyer settled 50 cents on a 5-cent listing because the payment header changed the price, a replayed payment number passed verification five times but settled once, a public cache on a paid endpoint let later buyers get data free, and an injected seller reply made a Llama 3.2 agent pay again (55 cents instead of 5).

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

Watch: AI Engineer: Your Agent Just Paid for Nothing: 6 Attacks on x402 Payments -…

Cohere study: agent teams gained up to 10 points on ProgramBench at 3-7x inference cost

On 200 ProgramBench tasks, Cohere reported teams improved reward by up to about 10 percentage points at three to seven times the inference cost; delegating only planning matched the best full team at about half the cost.

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

Watch: Cohere: Release Recap: Scaling Inference-Time Compute with Teams of Coding…

Models and learning

Nate Herk's 100-round improvement loop bloated and broke a site on Opus 5.5; Sonnet 5.5 stayed smaller

Nate Herk reported running the same 'ship a bold change' prompt 100 times each with Sonnet 5.5 and Opus 5.5 at medium effort in Claude Code.

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

Watch: Nate Herk: Make It Better Reel FINAL

How I AI: four agents given one launch-video prompt; Claude judged most complete, Grok next

A How I AI host gave four agents (Dot, Grok, Muse and Claude, per captions) the same ChatPRD launch-video prompt.

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

Watch: How I AI: 4 agents. Same ChatPRD launch video prompt

Fahd Mirza tests Phonon-2, a 164 MB English ASR model, on CPU

Per its model card as read by Fahd Mirza, Fermion Research's Phonon 2 averages 5.21% word error rate across seven English Open ASR sets, is built on Nvidia's Parakeet and is about 15 times smaller as a download.

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

Watch: Fahd Mirza: Phonon-2: 164MB Accurate Speech Recognition Model: Run Locally

Fahd Mirza tests Sage-A1.5, a 35B mixture-of-experts agentic model, locally in Hermes

Fahd Mirza said Sage A1.5 is a 35B mixture-of-experts model from a Shanghai lab and read a model-card chart as edging peers such as GLM 5.3.

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

Watch: Fahd Mirza: Sage‑A1.5: Agentic Orchestration, Verification, and Open Scientific…

Manolo Remiddi measures Gemma 4 and Qwen 3.5 speeds on a sponsored mini PC

On a sponsored mini PC, Manolo Remiddi reported Gemma 4 E4B at 16.87 tokens/s in LM Studio, Gemma 4 12B at about 9.4 to 9.7 tokens/s, and a Qwen 3.5 35B mixture-of-experts at 18.8 to 19 tokens/s.

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

Watch: Manolo Remiddi: AI Without a Subscription: Can a Mini PC Deliver?

Sam Witteveen reports HP ZGX Fury exceeds 3,000 tokens/s aggregate on Qwen 3.8 Flash Next

Sam Witteveen reported over 3,000 tokens/s aggregate, over 1,300 tokens/s with 16 concurrent users on a RAG test and about 88 to 90 tokens/s with 64 users on HP's ZGX Fury, a rack-mountable GB300 workstation with 748 GB shared memory (252 GB HBM at 7.1 TB/s), 20 petaflops at FP4 and two 400 Gb ports.

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

Watch: Sam Witteveen: 3,000 Tokens/Sec Locally: Testing the ZGX Fury

RidgeRun and Connect Tech shrink vision-language and robot-memory stacks to fit Jetson Orin NX 16 GB

In NVIDIA Developer sessions, RidgeRun said an 8B vision-language model's out-of-box vLLM allocation of 108 GB fell to 26 GB with a single-request KV cache and to 7.5 GB with TensorRT, INT4 language and FP16 vision parts, with 3.7 seconds per defect analysis.

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

Watch: NVIDIA Developer: How to Maximize Memory Efficiency and Reduce Costs on NVIDIA Jetson

Microsoft shows open-source Intelligent Terminal and Qwen 3.5 122B running on a Surface RTX Spark

Microsoft presenters said Intelligent Terminal is an open-source Windows Terminal fork with an ACP-compatible agent pane that works with bring-your-own models and harnesses, installable from the store, GitHub or winget.

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

Watch: Microsoft Developer: Running a 120B model offline on Surface RTX Spark Dev Box with…

Coverage

96 videos with transcripts reviewed (all fully read), 462 inventoried in the Oct. 27 to Oct. 31 scan window; all 96 fall in the 24-hour window and there are no late finds. Not reviewed: 58 under 45 seconds, 36 live or upcoming streams deferred, and 3 without transcripts. 13 new batches; 305 videos were already extracted in earlier editions. News: 2,149 articles in the window from 220 of 428 feeds. 290 feeds returned items, 2 failed and 14 were skipped as dead.

Sources reviewed (96 videos)

All 96 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.