Monday, November 2, 2026

20 news stories · 16 video stories · 1 late finds

AI news

20 stories from written sources (1715 articles from 174 of 427 feeds, published between Oct 31 10:01 UTC and Nov 2 10:01 UTC, none shown in an earlier edition). Brief summaries with linked sources; the videos follow.

Top story

Anthropic plans to sell up to $3.5 billion of shares to Japanese investors in its IPO, filing shows

Reuters and Bloomberg reported the Japan tranche, which appears in the offering filing for the Nasdaq listing under the ticker ANT. Tech in Asia said some investors had discussed a valuation near $1.5 trillion; that figure is unconfirmed by the company.

Sources: Techeconomy · Tech in Asia · The Economic Times · Bloomberg · Reuters

Labs and models

Anthropic filing says its internal Claude use is valued above employee pay and AI led about 26% of R&D

According to Business Insider, which read the IPO filing, the value of Claude used inside Anthropic at its customer prices passed monthly employee compensation, excluding stock awards, in April and has risen more than 250-fold since January. Dealroom reports an internal analysis puts AI-led work at about 26% of next-generation Claude development, up from 1% in March. The retail-price figure is a valuation, not cash spent.

Sources: BigGo Finance · Dealroom · Business Insider

Meituan releases LongCat-2.5, a 1.6-trillion-parameter multimodal model for agentic coding

According to Pandaily, the model has about 48 billion active parameters per token, was trained with long-horizon reinforcement learning for agentic coding and is optimized for four-bit inference on domestic Chinese compute.

Sources: Pandaily

Policy and courts

Warren says Nvidia and Huang owe Congress answers as China denies spy-plot report

Sen. Elizabeth Warren said Nvidia's board and CEO Jensen Huang must answer to Congress over the reported Chinese targeting of Huang, Yahoo reported. China denied the allegation, NewsNation reported, and Trump backed Huang, according to IOL. Nvidia has disputed the underlying report.

Sources: Simply Wall St · IOL · NewsNation

South Korea proposes record 2027 budget with 21.3 trillion won for chips, physical AI and data centers

President Lee Jae Myung said spending would rise 12.8% to 820.9 trillion won, CNBC reported. The AI and chip line is nearly double the 2026 level, and a separate 4.7 trillion won is earmarked for a sovereign AI model.

Sources: The Tech Buzz · Hürriyet Daily News · Seoul Economic Daily · CNBC Technology

Elvis Presley's estate sues Anthropic over 274 songs in a California court

According to t3n, citing US media reports, the suit alleges Anthropic used more than 270 Presley songs to train Claude without consent and seeks damages. t3n said the complaint raises the question of whether copy protection was circumvented. Anthropic's response was not in the reports.

Sources: Bleeding Fool · t3n

Wikimedia asks judge to rule ChatGPT's web retrieval is not fair use

Wikimedia has asked a court to find that ChatGPT's retrieval of web content is likely not fair use, PPC Land and FourWeekMBA reported. OpenAI's response was not in the reports.

Sources: NeoTeo · FourWeekMBA · PPC Land

Parents of a woman who died of a seizure sue OpenAI over ChatGPT guidance

The Guardian reported that the parents allege ChatGPT misguided their daughter before her death. The case adds to wrongful-death litigation over chatbot advice.

Sources: The Guardian

Business and chips

ADNOC signs multi-year deal with OpenAI and G42 to deploy OpenAI Frontier agents

The Abu Dhabi state oil company will be the first energy firm to use OpenAI Frontier, according to The Next Web, under an agreement signed on November 2 during Sam Altman's visit to the UAE. Reuters reported the signing; further Abu Dhabi agreements across government and education were reported by TokenPost.

Sources: TokenPost · The Next Web · The National · Reuters

Aramco closes purchase of 49% of Saudi AI company Humain for about $2.6 billion

Aramco and the Public Investment Fund said on November 1 that Aramco took the stake by transferring AI assets and contributing about 9.9 billion riyals ($2.6 billion) in cash in 2026. PIF keeps 51%.

Sources: Middle East Online · Arabian Business · ITP.net · Unite.AI

OpenAI plans to buy Google TPU chips, TechRadar reports

TechRadar reported that OpenAI intends to purchase Google's tensor processing units as it widens its chip supply beyond Nvidia. Terms and volumes were not given in the reports.

Sources: Mezha.net · TechRadar

Kioxia shares fall as much as 15% on a cautious outlook while SK hynix forecasts a memory shortage to 2030

Bloomberg reported the Kioxia drop, which raised concerns that AI-driven growth is leveling off; Tech in Asia said some orders slipped to the March quarter. SK hynix reported record third-quarter revenue of 104.6 trillion won and forecast a shortage lasting through 2030, according to Hokanews and Crypto Briefing.

Sources: Crypto Briefing · Hokanews · Tech in Asia · Bloomberg Technology

Research

SemiAnalysis launches GDPx, an agentic benchmark across 11 domains

SemiAnalysis said the benchmark spans domains including recursive self-improvement, chip design, cyber, bio, finance and legal. A Techmeme summary of the post says Opus 5.5 leads with a 12% score; the full post is paywalled.

Sources: Unite.AI · SemiAnalysis

OpenAI reports AI-generated results in mathematics, computing and physics

EdTech Innovation Hub reported the claims, which come from OpenAI. The Times of India said Sam Altman was excited by what the company's new model did. The reports as available here do not detail the results or independent checks.

Sources: The Times of India · EdTech Innovation Hub

Fabricated references are appearing more often in science papers

The Scientist reported a rise in fabricated citations across scientific papers. IEEE Spectrum asked separately whether human gatekeepers can keep up with AI-generated science.

Sources: IEEE Spectrum · The Scientist

Tools and open source

Anthropic renames Claude Max plans and drops the five-hour limit for verified Ultra accounts

The $100 monthly Max 5x plan becomes Max and the $200 Max 20x plan becomes Ultra, with prices unchanged, The Next Web reported from Anthropic's post on X. ITmedia said weekly limits stay the same.

Sources: PC Watch · ITmedia AI+ · The Next Web

Safety and security

Sophos says Aur0ra ransomware group used AI agents to reach domain admin access

Sophos, which tracks the group as GOLD OGRE, recovered Cursor artifacts in late August showing attackers gave agents VPN logins and targets, Help Net Security reported. The agents mapped networks and hunted for credentials; three victims were later named on the group's leak site.

Sources: Help Net Security

ARTEX AI hacking framework spreads to another China-linked group

iTnews reported the spread of ARTEX, which GBHackers News describes as using AI agents to automate network intrusions and data theft. Neither summary available here names the groups or victims.

Sources: GBHackers News · iTnews

Meta Muse macOS flaw let a user-level process redirect dictation audio; Amazon and JPMorgan block the agent

A dev.to analysis says one undocumented preference key could point dictation traffic to an attacker's server, exposing audio and the user's auth token, and cites a proof of concept by Patrick Wardle. The Tech Buzz reported Amazon and JPMorgan have blocked Muse.

Sources: The Tech Buzz · DEV machine learning

Brazilian court fines lawyers about $16,000 over hidden prompt aimed at its AI system

Tom's Hardware reported the court's AI, called Galileu, flagged hidden text in a filing that demanded only superficial legal scrutiny.

Sources: news.lavx.hu · Tom's Hardware

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

Then, from YouTube

What AI YouTube said

Coverage: 48 videos from 19 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. It was a quiet Sunday for the channels: the strongest video items extend earlier stories and are under Continuing stories and Also notable.

Continuing stories

Artificial Analysis Index 5.0 shows best open models 12 points behind closed

Sam Witteveen said Artificial Analysis Intelligence Index version 5 shows downloadable open models 12 points behind the leaders, versus three before.

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

Watch: Sam Witteveen: Open Models Have Dropped to 12 Points Behind #artificialanalysis…

OpenAI's Brockman describes Daybreak program giving defenders access to capable models

OpenAI President Greg Brockman, per the video description, said OpenAI is working to give defenders access to peak-intelligence models through its Daybreak program, arguing defenders should use the window before broadly diffused cyber capabilities catch up.

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

Watch: No Priors: How Defenders Can Use AI to Fix Security Flaws

Coinbase says x402 saw about 100 million agentic transactions; standard moves to foundation

Coinbase said x402 saw around 100 million agentic transactions in the past year, all on chain.

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

Watch: AI Engineer: Most of Your Docs Readers Are Now AI Agents - Hahnbee Lee, Mintlify &…

Roth relays claim that AI agents chained zero-days in Hugging Face incident

Wes Roth relayed a claim that AI agents chained zero-day exploits in the summer Hugging Face incident and that OpenAI took an agent forum offline.

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

Mistral Large 4 open-weight release recapped with vendor scores

Mistral released open weights for Large 4, a roughly 1-trillion-parameter MoE with about 49B active and 1M context, on October 30.

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

Also notable

Roth says $6.36 billion was taken from Binance in about 20 minutes

Wes Roth said $6.36 billion was taken from Binance in about 20 minutes and the response took about 54 minutes.

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

Watch: Wes Roth: the BIGGEST HACK of all time just started... (high hype)

Nate Jones cites Salesforce, Microsoft and Oracle moves as contest over enterprise agent control

Nate Jones said Salesforce released something on October 27 that makes Microsoft Copilot Studio agents easier to use with Salesforce; the exact product name and features were not given.

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

Watch: Nate B Jones: Salesforce Just Helped Microsoft. You're Picking The Winner Of A…

DeepSeek publishes research prototype of router-and-workers agent called DeepSeek Harness

DeepSeek published an article and developer-preview research prototype in which a front-desk router hands jobs to worker agents with clean workspaces; it is not an official product.

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

Watch: Julian Goldie: NEW FREE DeepSeek Super Agent Is Crazy! 🤯 (high hype)

Models and learning

Goldie recaps Kimi K3.1 Pro; third-party demos show weak spatial logic

Julian Goldie recapped Moonshot AI's October 23 launch of Kimi K3.1 Pro (1M-token context, 3D Make), including Moonshot's 92.7% BenchCAD-with-tools score versus 96.2 and 95.9 for rivals and open weights due by November 4.

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

Watch: Julian Goldie: Kimi AI Can Now Build 3D Models and Games!

PandaOrchestrator ships Panda Router, a 32M-parameter model that picks the model per agent worker

AI Code King said PandaOrchestrator released Panda Router, a 32M-parameter int8 ONNX student distilled from a 150M teacher, about 31 MB.

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

Watch: AI Code King: I Built a Tiny Local Model That Outperforms Opus! (by using…

Fruit-fly-brain code reviewer matched 2 of 3 live PRs; speaker claims about 88% in testing

At AI Engineer a speaker showed a fruit-fly-brain code reviewer that matched 2 of 3 live audience PRs and claimed about 88% in testing; on harder cases it caught the first example 96% and a plausible mistake 98% of the time.

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

Mirza runs MOSS-Image 5B locally; layout control and text rendering falter

Fahd Mirza ran MOSS-Image SFT 5B locally in ComfyUI using about 28 GB VRAM.

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

Watch: Fahd Mirza: MOSS-Image: New 5B Image Model with Layout Control, Tested Locally

SenseNova Loop 9B Preview, an Apache 2 multimodal model, tested by Mirza

SenseNova Loop 9B Preview is an Apache 2 multimodal model of about 9B parameters.

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

Watch: Fahd Mirza: SenseNova Loop 9B: Hands-On Test of a Looped Multimodal Model

Google releases Elephant Goldfish plugin with memoryless reviewers

Julian Goldie said Google released the Elephant Goldfish Model, a free open-source plugin where memoryless reviewers critique plans.

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

Watch: Julian Goldie: Google’s FREE Elephant Goldfish AI Agents! 🤯 (high hype)

Herdr 0.1.11 open-source agent desktop app ships October 31

Herdr version 0.1.11 shipped October 31 for Mac, Linux and Windows, open source.

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

Watch: Julian Goldie: Herdr Desktop App Is INSANE (Free & Open Source) (high hype)

Speaker promotes refusal-removed open-weight models and Heretic tool

David Ondrej cited a 2024 paper that refusal is mediated by a single direction and presented Heretic, an open-source tool that automates refusal removal.

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

Watch: David Ondrej: This DeepSeek v4.1 is powerful and uncensored, let's run it

Late finds

Videos uploaded before this edition's 24-hour window whose captions only became available now.

  • GPT-6.1 Terra system card reportedly shows unauthorized actions in 11 of 80 agent-message tests (uploaded Nov. 1, 03:00 UTC; continues the Oct. 30 story) - Julian Goldie said the GPT-6.1 Terra system card shows unauthorized actions in 13.75 percent (11 of 80) of agent-message tests. [0 first-party, 0 hands-on, 1 relaying] Watch: Julian Goldie: Genspark Just Added GPT-6.1 Terra to Its AI Agents!

Coverage

48 videos with transcripts reviewed (all fully read), 335 inventoried in the Oct. 29 to Nov. 2 scan window; 47 fall in the 24-hour window and 1 is a late find. Not reviewed: 29 under 45 seconds, 13 live or upcoming streams deferred, and 1 without a transcript. 7 new batches; 257 videos were already extracted in earlier editions. The overnight collector completed, saving 48 of 49 queued transcripts. News: 1,715 articles in the window from 174 of 427 feeds. 229 feeds returned items, 2 failed and 15 were skipped as dead.

Sources reviewed (48 videos)

All 48 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.