Wednesday, November 11, 2026

17 news stories · YouTube digest in a later edition

AI news

17 stories from written sources (264 articles from 71 of 424 feeds, published between Nov 10 10:00 UTC and Nov 11 14:57 UTC, none shown in an earlier edition). Brief summaries with linked sources; the videos follow.

Top story

Google makes Gemini 4 Carbon available to developers, enterprises, scientists and consumers

Google DeepMind said on Nov. 10 that Carbon is its strongest model for cybersecurity and long-horizon agentic work in knowledge tasks, software engineering and scientific reasoning, following a limited release to security defenders. Google also launched the MKA benchmark, built with Artificial Analysis, to test agents on visually complex files.

Sources: Economic Times Tech · Google Gemini blog · Google DeepMind

Policy and courts

Nvidia hires Wachtell Lipton for Huang inquiry as Sen. Cramer urges him to resign and Trump says FBI is looking into the affair

Politico reported Nov. 10 that Nvidia retained Wachtell Lipton for an independent investigation, which NBC News said former acting US Attorney Joon Kim will lead. Sen. Kevin Cramer called Huang a national security risk, and Tom's Hardware reported three committee chairs sought a closed-door briefing by Nov. 11, with a subpoena possible.

Sources: The Hindu Technology · Tom's Hardware · NBC News Tech · Politico Technology

New York's AI safety law takes effect as Hochul launches a state AI testing program

Developers such as OpenAI and Anthropic must now register, disclose safety practices and report critical incidents or face fines under the law, Bing News reported Nov. 10. Politico said Gov. Kathy Hochul's program is the first of its kind among states and arrives amid federal inaction on AI rules.

Sources: Politico Technology

Democrats weigh AI bills including narrower state preemption and a federal permission regime for model training

Politico reported Nov. 10 that Rep. Lori Trahan is drafting an AI safety bill that would preempt fewer state laws. Futurism reported a draft bill would create a Department of Artificial Intelligence with authority over training, and that Rep. Yassamin Ansari urged a pause on advanced AI development.

Sources: Futurism · Politico Technology

Mistral urges French lawmakers to reject a proposed AI copyright law, drawing criticism from creatives

Sifted reported Nov. 10 that the Paris-based model maker is asking parliamentarians to vote against a proposal on the use of copyrighted material in generative AI. Writers, journalists, musicians and artists in France have criticized the company's stance.

Sources: Sifted

Business and chips

Anthropic seeks IPO valuation of $1.3 trillion to $1.4 trillion as marketing begins, outlets report

Anthropic said in a filing on Nov. 10 that it will sell 825 million shares at $70 to $78, ABC News reported, which would raise more than $61 billion at the midpoint. ABC put the valuation target near $1.3 trillion; NBC News and the New York Times reported up to $1.4 trillion. Economic Times cited a $1.02 trillion valuation.

Sources: Economic Times Tech · New York Times Technology · NBC News Tech · ABC News Technology · Anthropic

OpenAI's third-quarter revenue rises about 50% to $10 billion, with an adjusted operating margin of -109%, WSJ reports

The Wall Street Journal reported Nov. 10, as relayed by Reuters, that OpenAI told investors its adjusted operating margin improved from -150% in the prior quarter. Operating losses widened to $10.9 billion, according to documents the Journal viewed. OpenAI did not immediately comment.

Sources: The Hindu Technology

TSMC board approves up to $500 million each in Anthropic and OpenAI

TSMC said in a statement that the investments in Anthropic common shares, at its IPO price, and OpenAI preferred stock are capped at $500 million each, or $1 billion in all, according to Economic Times.

Sources: Economic Times Tech

CoreWeave beats third-quarter revenue estimates at $3.66 billion and raises its annual forecast

Revenue rose from $1.37 billion a year earlier, against an average analyst estimate of $3.55 billion from LSEG, Economic Times reported. CoreWeave raised its 2026 forecast to $13.0 billion to $13.3 billion, and shares rose more than 7% in extended trading after a slump of more than 40% in the past month.

Sources: Economic Times Tech

AMD begins shipping Zen 6 EPYC SP7 server CPUs and disputes rivals' 'agentic CPU' claims

AMD said Nov. 10 that sixth-generation EPYC SP7 systems are shipping from server makers, with Azure and OCI instances coming soon and Helion AI racks rolling out this month. Tom's Hardware reported AMD executives pushed back on Nvidia and Arm marketing of CPUs built for agentic AI.

Sources: Tom's Hardware · Data Center Knowledge · AMD IR

Research

Anthropic alignment researchers report diluting poisoned fine-tuning data can strengthen its effect

According to the title of a post on Anthropic's Alignment Science blog dated Nov. 10, mixing poisoned data into larger fine-tuning sets can make the poisoning more effective rather than weaker. Only the title was available, so methods and figures are not confirmed here.

Sources: Anthropic Alignment Science blog

Tools and open source

OpenAI launches Businesses in ChatGPT, letting companies claim profiles and deploy sales agents

OpenAI said Nov. 10 the beta lets verified companies edit details, add product feeds and see aggregated views and clicks. Business Agents, initially for eligible US e-commerce and online service firms, answer customer questions. OpenAI said it does not share user chats with businesses.

Sources: Economic Times Tech · OpenAI News

Anthropic adds spreadsheet, whiteboard and dashboard tools to Claude

Economic Times reported that Anthropic announced on Nov. 10 new Claude tools for creating and editing spreadsheets, collaborating on whiteboards and generating data dashboards. Claude Sheets supports collaborative formulas and data work, putting the product alongside Microsoft 365 and Google Workspace.

Sources: Economic Times Tech

Safety and security

Researcher finds more than 35,000 Grok Build apps open to anyone, some exposing passwords and company data

Ars Technica reported Nov. 10 that share links for apps built with SpaceXSI's Grok Build can be enumerated despite randomized URLs, exposing data tied to each project. The leaked material includes proprietary business information and some of the company's own, according to research the outlet cited.

Sources: Ars Technica AI

SGLang flaw exploited to breach a NEAR AI inference server, exposing data from 18 agents and about 100 users

According to The Hacker News, an attacker used a remote code execution flaw in the open-source SGLang engine to run code on a NEAR AI server last Sunday. NEAR AI said the attacker could also have seen requests processed that day, including those sent in confidential mode.

Sources: The Hacker News

Microsoft researchers say fewer than 10 neurons can be edited to plant a backdoor in an AI agent's model

According to The Hacker News, Microsoft researchers showed that editing fewer than 10 rarely used neurons, without retraining, can make a model emit an attacker's command when a chosen trigger appears. An agent would run that command with its own permissions, the report said.

Sources: The Hacker News

Frontier Model Forum publishes a framework for grading the severity of AI jailbreaks

Tom's Hardware reported that the Jailbreak Rating Framework works like the CVSS scheme for software flaws, rating jailbreaks by capability gained, scope and ease of deployment. The forum was founded in 2023 by Anthropic, Google, Microsoft and OpenAI.

Sources: Tom's Hardware

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

Then, from YouTube

What AI YouTube said

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.

YouTube digest

Not built on this day. The transcripts of the uploads listed below, where available, were read for the Nov. 13 edition, where their stories appear as late finds.

Sources reviewed (129 videos)

129 videos uploaded in the 24 hours to 03:13 UTC on Nov 11, grouped by channel. Their transcripts, where available, were read for the Nov. 13 edition.