Get in touch

A.I. news of Wednesday 11 November 2026

Archive From the archive of The Latest A.I. News

This is an archive page with the A.I. news of Wednesday 11 November 2026, as it stood that day. The summaries have not been updated since, and a lot may have changed. Today's news is on The Latest A.I. News.

Last updated:

Today is dominated by Anthropic's IPO (now also open to Australian retail investors), its new Claude work apps, Gemini 4 Carbon and the chip-export row around Nvidia and China.

Update: Anthropic starts its IPO roadshow: 825 million shares at $70-78, to list as 'ANT' on Nasdaq

Certain Business

Anthropic has officially announced that its roadshow has begun. It is offering 825 million shares, more than 801 million of them newly issued by the company, at an expected $70 to $78, with Goldman Sachs and Morgan Stanley as banks. That differs slightly from the earlier Reuters report (867 million shares, $70-80); the registration is not yet effective, so final terms may still change.

Update: Anthropic also opens an Australian retail offering; closes 17 November, listing on Nasdaq only

High Business

From Wednesday morning, Australian retail investors can subscribe to Anthropic shares through eight brokers led by CommSec. The offer closes on 17 November, the price range stays at $70-78, and the stock will trade as 'ANT' on Nasdaq only, not on the ASX. It is part of the IPO of up to $64.4 billion plus a $26 billion private placement.

Google makes Gemini 4 Carbon widely available, with an intro price below Claude Fable 5.2

High Models Products

About two weeks after the announcement, Gemini 4 Carbon is available to Ultra subscribers in the Gemini app, AI Mode and Antigravity, and through the paid API. Intro pricing is $2 input and $10 output per million tokens until November 30, then double. VentureBeat's analysis says Carbon scores below Fable 5.2 on the Artificial Analysis index, and some of Google's claims lack published comparison scores.

Anthropic unveils Claude Sheets, Meetings and Whiteboards; Claude Data and Interactive Connectors now generally available

High Products

At an event in San Francisco, Anthropic showed Claude Sheets (beta on all paid plans, with Excel export), Meetings (Team/Enterprise only, joins Google Meet calls) and Whiteboards. Claude Data, the successor to Dashboards, and Interactive Connectors are generally available, with plugins for Adobe, Atlassian and HubSpot among others. This confirms the earlier rumour.

OpenAI lets businesses claim a ChatGPT profile and attach their own sales agent (beta)

High Products Business

Verified businesses can manage their own listing in ChatGPT and deploy an agent that answers customers and sells. The beta starts in the US with ecommerce and online service businesses. It fits OpenAI's push to position ChatGPT as a search and shopping channel too.

Google DeepMind unveils Gemini Robotics 2.5; humanoid reaches 96% on a Mercedes line

Certain Models Research Safety

DeepMind says the model succeeds 30 to 100% more often than its predecessor across more than 200 tasks and works 20 to 30% faster. In a pilot at Mercedes-Benz, Apptronik's Apollo 2 humanoid reached 96% success on parts kitting. The model is rolling out to robotics partners only for now; DeepMind also released an open safety benchmark against jailbreaks.

TSMC board approves up to $500 million for Anthropic shares at the IPO price and $500 million for OpenAI

Certain Business Chips

Board resolutions published with quarterly results say TSMC may invest up to $500 million in Anthropic shares at its IPO price and up to $500 million in OpenAI preferred stock. A further $500 million is approved for a new strategic investment company. It pulls the chipmaker deeper into financing the big AI labs.

Update: Nvidia affair continues: PR firm, an earlier short-seller warning and a race to sue Nvidia

Medium Policy Chips Business

According to TechCrunch, Huang had hired a PR firm experienced in sex scandals a year ago, and a short seller flagged his ties to the alleged Chinese spy six months before the scandal broke. Lawyers are reportedly preparing claims against Nvidia, and Huang pulled out of the ElevenLabs Summit. This is reporting from one outlet; Nvidia's own inquiry is still under way.

Chinese university bought two Nvidia H200 chips despite export ban; Congress pushes chip bills toward the NDAA

Medium Chips Policy

Purchase records show Northwestern Polytechnical University bought two H200 chips on 21 October via a reseller for about 874,700 yuan (roughly $123,000), with no Nvidia involvement or export licence; Reuters flagged 26 similar records. After the Nvidia affair, US lawmakers want to attach the AI Overwatch Act, the Chip Security Act and the Match Act to the NDAA.

Google DeepMind unveils AlphaProof Code: AI verifies cryptographic guarantees in Signal's protocol code

Certain Research Models

AlphaProof Code, built on Gemini Carbon, wrote over 6 million lines of Lean in about four weeks to formally prove selected cryptographic guarantees in Signal's protocol code. Around 1% was flagged for human review, and two previously fixed bugs that had been reintroduced were caught. The specification and proofs have been open-sourced.

New York opens frontier-AI registration portal: OpenAI and Anthropic must register by 1 January

Certain Policy Safety

Under New York's new AI safety law, developers with more than $500 million in revenue must register by 1 January 2027, submit safety frameworks, report critical incidents within 72 hours and publish transparency reports. Fines go up to $1 million for a first violation and $3 million for repeats.

Leaked figures: OpenAI posted a $10.9 billion Q3 operating loss on about $10 billion revenue, per the WSJ

Medium Business

According to the Wall Street Journal, OpenAI's operating loss grew from $10.1 to $10.9 billion while revenue rose about 50% to roughly $10 billion and the adjusted margin improved from -150% to -109%. The same report cites leaked Anthropic figures of $17 billion revenue and $2.5 billion adjusted operating income; not officially confirmed.

Update: Meta's Muse Ember nearly matches Claude Haiku 5.5 at a third of the cost per task, says VentureBeat

Medium Models Business

VentureBeat compares Meta's new proprietary Muse Ember with Anthropic's Haiku 5.5 and estimates it performs almost as well at roughly a third of the per-task cost. Meta earlier presented Ember as a cheap model with a one-million-token context. The comparison rests on the outlet's estimates, not an official benchmark.

Anthropic filing: Lambda missed a lease deadline, $34.9 billion of compute commitments terminable

Medium Business Chips

The amended S-1 says Lambda missed a lease deadline for Hut 8's Beacon Point site in Texas, making $34.9 billion of commitments terminable; Anthropic says it has no plans to walk away. Nvidia and Google each take about $10 billion of non-voting stock at the midpoint of the price range.

VulnCheck: Anthropic's Project Glasswing yielded 442 CVEs, but accuracy falls short

Medium Safety Research

VulnCheck attributed 442 CVEs (0.5% of 2026's) to Glasswing, of which only 2 are known to be exploited. Of 37,966 self-reported findings, 7,349 are public and 217 fixed; curl maintainer Daniel Stenberg rejected 19 of 34 reports. VulnCheck measures a 55.9% false-positive rate versus Anthropic's claimed 94.2% accuracy.

Microsoft Research presents 'Lazarus': backdooring AI agents by reprogramming fewer than 10 dormant neurons

Certain Research Safety

Microsoft researchers show an AI agent can be backdoored by modifying fewer than ten dormant neurons, without any retraining. The team used Claude Code and Opus 4.5 in the discovery process.

Update: Notion also hit by the Fireworks AI data leak; AI chat transcripts of under 1% of workspaces exposed

Medium Safety Business

Notion says Notion AI chat transcripts used for Insights were exposed at third-party vendor Fireworks AI. Notion disconnected on 3 November, was notified on 6 November and confirmed the dataset on 8 November. Users are told to rotate credentials they pasted into Notion AI.

Security firm: thousands of Grok Build apps were open without login, some with private data

Low Safety

Claroc found 35,578 live Grok Build apps accessible without login; an estimated 650 to 800 held private data, including 24 belonging to SpaceXSI and X staff. Affected apps were taken down after disclosure. Not yet confirmed by xAI.

Survey: companies blame stalled AI projects on cost overruns three times as often as on compute shortages

Medium Business Research

VentureBeat reports from its own survey that cost overruns are cited three times as often as a lack of compute as the reason AI projects stall. That fits earlier signals such as Ramp's data showing corporate AI spending flattening.

PayPal: AI assistant jumps from 43 to 95% accuracy without changing the model, thanks to a knowledge layer

Medium Research Products

PayPal says its internal data assistant became far more accurate thanks to a separate layer of tested example queries and analysis playbooks, while the underlying language model stayed the same. It cites over 1,000 questions and about 1,600 hours saved in ten weeks. These are PayPal's own figures, not independently verified.

Stanford AI100 report: the scaling era is hitting limits of data and compute

Medium Research

The third edition of Stanford's AI100 report, chaired by Michael Wooldridge, says the scaling era is running into limits of data and compute. It also warns of coordinated cyberattacks by AI agents on infrastructure.

ElevenLabs releases Scribe v3 and Scribe v3 Live for speech recognition, with real-time speaker identification

Low Models Products

According to a daily digest there is a version for recorded audio and a real-time one, with fewer errors and live speaker separation. The error reductions cited come from the company itself and are not yet independently confirmed.

UK government: AI software on 54% of compatible MRI scanners, services doing about 14% more scans

Certain Products Other

NHS England and the health department say AI acceleration software runs on just over half of compatible MRI scanners. Services using it reportedly perform about 14% more scans. These are government figures without independent evaluation.

PwC launches a 'critical thinking' certificate for more than 85,000 staff in the AI era

Certain Business Other

The firm rolls out a programme from December 2026 and aims for completion by June 2027. In early 2027 PwC plans to share the framework with universities. It is one of the first large employers to formalize skills for judging AI output on such a scale.

Next up: 17 November: Australian offer closes; Anthropic's Nasdaq debut possible in the week of 16 November

All A.I. news

The short version

Wednesday 11 November 2026

  • Anthropic presses on with its IPO: Australian retail offer open until 17 November ($70-78, 'ANT' on Nasdaq), while its filing flags a risk around a Lambda deal.
  • Anthropic officially unveiled Claude Sheets, Meetings and Whiteboards; Google made Gemini 4 Carbon broadly available at an introductory price below Claude Fable 5.2.
  • The Nvidia row continues: records show a Chinese university bought H200 chips despite the ban, and US lawmakers want chip bills added to the NDAA.
  • Also: New York opened its frontier-AI registration portal, the WSJ reports a $10.9 billion Q3 loss at OpenAI, and DeepMind showed AlphaProof Code.

What does TL;DR mean?

TL;DR stands for “Too Long; Didn’t Read”. It’s the name for a short summary for people who are short on time. Click the button for the key points of this day in a few lines; all of the day’s items are right below.

Archive

Every day has its own page. They stay up for three months.

November 2026

October 2026

September 2026

Everything here is a summary of reports from other media, not original research. The label (Certain, High, Medium, Low) is my estimate of how reliable the source is. The summaries are generated automatically with AI and may contain mistakes; read the article itself for the details (the star marks the leading source). Sources: leading A.I. and tech sites and the official blogs of the A.I. companies.