Google ties school Search access to AI

Google says Search and Gemini will be off by default for K-12 student accounts unless administrators turn them on, and critics call it coercive. Also inside: Bloomberg reports OpenAI raised $11 billion, METR measures Anthropic's pace of improvement, and AI stocks fell after the midterms.

AI lab releases

Google ties schools' access to Search to its AI features

Google says Gemini and Google Search will be off by default for K-12 student accounts next year unless administrators turn them on. Chris Phillips, who leads education at Google, told The New York Times that schools blocking AI will lose Search as well. Critics including Senator Bernie Sanders and former FTC chair Lina Khan called the move coercive, while Google says schools remain in control.

Why it matters: Schools that rely on Chromebooks and Google tools must now choose between keeping Search and keeping AI out.

Sources: TechCrunch, The Verge, The Decoder, SiliconANGLE, The Information

Google's Gemini Omni 1.2 Flash makes longer AI videos for less

Google has made Gemini Omni 1.2 Flash, its video-generation model, generally available, meaning open to all customers rather than in testing. Clips can now run up to 15 seconds, up from 10, and the price falls 30% to $0.07 per second of 720p video. Google says users can refine a clip with follow-up instructions instead of starting over, and claims faces and products stay more consistent.

Why it matters: Cheaper, longer clips lower the cost for marketers and developers trying AI video in ads and products.

Sources: Google DeepMind, Google, The Decoder

Thinking Machines releases Inkling 2 Small, its second model generation

Thinking Machines, founded by former OpenAI CTO Mira Murati, released Inkling 2 Small and will publish its weights, the downloadable model files, openly. The company says the model scores 31 on the Artificial Analysis Intelligence Index, close to Google's Gemini 3.8 Flash. Business Insider reports the startup is raising $5 billion in a round led by Accel at a $40 billion valuation.

Why it matters: Developers who want to fine-tune a model they can run themselves get another option from a US lab.

Sources: The Decoder, SiliconANGLE

Claude can now read your bank accounts to explain spending

Anthropic launched Personal Finance, a Claude feature that connects to bank and card accounts through Plaid, a service that links banks to apps. Anthropic says access is read-only, so Claude cannot move or invest money, and finance chats will not be used for training. The beta is rolling out to US Pro, Max and Ultra subscribers, months after OpenAI added a similar ChatGPT feature.

Why it matters: People weighing whether to share bank data with a chatbot now have two major options, each limited in what it can see.

Sources: TechCrunch, The Verge, The Decoder, The Information

Acquisitions + deals

Bloomberg: OpenAI raises $11 billion at $1.25 trillion valuation

Bloomberg reports OpenAI has raised $11 billion in a new round that values the company at $1.25 trillion. Bloomberg says the round could total about $20 billion when it closes in the coming weeks. The Verge and The Decoder say OpenAI first aimed for $30 billion at a $1.4 trillion valuation before AI stocks fell.

Why it matters: Investor appetite for the largest AI labs appears to be cooling, which could shape plans to go public.

Sources: TechCrunch, The Verge, The Decoder, The Information

Meta buys Conductor, a tool for running coding agents in parallel

Meta has acquired Conductor, whose app lets developers run several AI coding agents, programs that write code on their own, at once. Conductor's cloud service shuts down on November 20, while its Mac app keeps getting updates until early 2027. The deal came on the same day OpenAI announced plans to buy developer tools startup Pierre Computer.

Why it matters: Big AI companies are buying developer tools to keep programmers on their own platforms.

Sources: TechCrunch, SiliconANGLE, Conductor, The Decoder

Bloomberg: Microsoft winding down Oracle AI cloud rental deal

Bloomberg reports Microsoft is winding down an arrangement to rent AI computing capacity from Oracle. Bloomberg says the move is part of a wider review of Microsoft's cloud partnerships worth billions of dollars.

Why it matters: A pullback by a major renter of capacity is a signal to watch for companies selling AI computing.

Sources: Bloomberg, The Verge, The Information

Agility Robotics begins trading on Nasdaq after SPAC merger

Agility Robotics, maker of the two-legged Digit robot, listed on Nasdaq by merging with a SPAC, a shell company that is already public. The deal raised about $465 million, and GeekWire says it valued Agility at $2.5 billion before the new money. TechCrunch says Agility earned $174,000 in revenue and lost $362 million in the first half of 2026.

Why it matters: Investors can now buy shares in a pure humanoid robot maker and follow its results in public filings.

Sources: TechCrunch, SiliconANGLE, GeekWire

More funding

• Crunchbase News reports General Catalyst led more US startup rounds than any other investor in October, worth $5.1 billion combined. Sources: Crunchbase News
• TechCrunch Venture reports former Secret Service agent Stephen Ward is raising a second cybersecurity fund of about $250 million at Brightmind Partners. Sources: TechCrunch
• Crunchbase News reports Siggy raised $6 million in pre-seed funding led by Village Global for an AI-assisted mental health service. Sources: Crunchbase News
• Crunchbase News reports construction robotics startups raised about $1.6 billion this year through October 29, led by Bedrock Robotics. Sources: Crunchbase News
• SiliconANGLE reports Ocient raised $100 million in a Series C to grow its platform for AI agents analyzing very large datasets. Sources: SiliconANGLE

MCP + agent tooling

Perplexity launches always-on AI agents called Teammates

Perplexity introduced Teammates, AI agents that run around the clock and keep working through ongoing projects. Each one runs on its own virtual machine, and users can follow progress in the Perplexity app or in Slack. Teammates need Perplexity's $200-a-month Max plan, while OpenAI's Dots start at $100 and Anthropic's Duet at $20.

Why it matters: Anyone comparing always-on work agents has another option, though at a higher entry price.

Sources: The Verge, The Decoder

Revolut makes a cheaper open model its default coding assistant

A Devsplainers video says Revolut made GLM 5.3, an open model from Chinese lab ZAI, the default coding AI for 2,000 engineers. David Abitbol, who runs AI operations at Revolut, says the bill is a tenth of what it paid Anthropic and OpenAI. GLM 5.3 scores well below Claude on coding tests, but Revolut reports faster delivery once engineers stopped rationing use.

Why it matters: Companies paying large AI coding bills may find unlimited use of a weaker model worth testing.

Sources: Devsplainers (YouTube), The Register

Box adds Nvidia models to help AI agents search company files

Box announced Box Content Retrieval, built on Nvidia's Nemotron embedding models, which map text by meaning rather than keywords. Nvidia and Box say queries run about four times faster and use up to 90% fewer tokens, the text units AI services bill by. Box says the feature follows existing file permissions and reaches Enterprise Advance customers early next year.

Why it matters: Businesses wary of agents reading sensitive files get a tool built around their existing access rules.

Sources: TechCrunch, SiliconANGLE

GitHub Copilot arrives in JetBrains Air and Xcode

GitHub's weekly update brings Copilot to JetBrains Air and to the coding assistant built into Apple's Xcode. The Xcode integration is in public preview, starting with the Xcode 27.3 beta. The Copilot app can now merge a pull request, a proposed code change, once it is ready.

Why it matters: Apple and JetBrains developers can use Copilot without leaving their usual tools.

Sources: The GitHub Blog

SDKs + APIs

Amazon SageMaker can now split one GPU among several models

Amazon SageMaker Inference now supports Nvidia's Multi-Instance GPU, which divides one GPU, the chip that runs AI models, into separate partitions. Each partition gets its own share of compute and memory, so different models can run side by side in isolation. The feature works only on endpoints that use inference components, not on single-model or serverless endpoints.

Why it matters: Teams running many small models may pay for less idle chip capacity.

Sources: Amazon Web Services

Web scraping + data tools

Playwright 1.65.0 lets its test runner work under Bun

Microsoft released Playwright version 1.65.0, and its test runner can now run under Bun instead of Node.js. Some features work differently under Bun. For example, there is no watch mode. A new reporter lists only failures. Breaking changes affect cookie matching, proxy credentials and TypeScript class fields.

Why it matters: Anyone maintaining Playwright tests should review the breaking changes before updating.

Sources: Microsoft on GitHub

Research

METR finds AI has roughly doubled Anthropic's pace of improvement

METR, an AI safety nonprofit, spent 11 weeks inside Anthropic and found its models are improving about twice as fast as before. The Decoder reports METR attributes about 80% of that speedup to AI being used in Anthropic's own research. Anthropic was allowed to redact parts of the report, and The Verge says it used that power liberally.

Why it matters: Safety researchers treat AI speeding up its own development as an early warning sign, so outside checks like this get close attention.

Sources: The Decoder, The Verge, The Information

Nvidia says its open 6G wireless tools are spreading in research

Nvidia says developer use of its open source Aerial and Sionna wireless software grew tenfold over the past year. The company says more than 8,000 research institutions use the tools to work on 6G, the next generation of mobile networks. Nvidia cites a King's College London project that cut a robot arm's task time by up to 52.6% using compact wireless messages.

Why it matters: Shared open tools could let 6G research move faster across labs.

Sources: NVIDIA Blog

ToyBlocks offers a simple simulated world for studying robot learning

The researchers behind MolmoAct 2 and MolmoBot released ToyBlocks, an open simulation of a robot arm handling blocks on a table. It creates training demonstrations automatically, so nobody has to steer the robot remotely to record them. The team calls it the first of several controlled environments, named FoundationWorlds, that it plans to build.

Why it matters: A controlled test bed could help researchers trace why robots fail, not only show what they can do.

Sources: Hugging Face

OpenAI launches a benchmark for routine lab data analysis

OpenAI introduced BioBasicsBench, a benchmark, or standard test, of 100 routine wet-lab tasks built with biotech company Ligo. Agents must read raw evidence such as gel images and microscopy, spot errors, and say when data cannot back a conclusion. In OpenAI's own testing, GPT-6 Astra earned 66.7% of available credit, compared with 43.8% for GPT-5.6 Sol.

Why it matters: Scientists considering AI help with lab work get a measure built on messy, real experimental data.

Sources: OpenAI

Other notable news

AI stocks fall after Democratic midterm wins as Trump blames voters

The Nasdaq fell more than 5% on Thursday as AI-linked stocks sold off after Democrats won control of Congress. The BBC reports several Democrats campaigned against data center growth, and investors fear more AI regulation. President Trump posted that voters had only themselves to blame for the market drop.

Why it matters: Policy risk is now part of how markets value AI companies and the data centers behind them.

Sources: Axios, BBC, Reuters

Lawmakers ask Jensen Huang to brief Congress on alleged spy ties

Three House Republican committee chairs asked Nvidia CEO Jensen Huang for a closed-door briefing on his contacts with Alice Huang. The New York Times reports China's intelligence ministry recruited Alice Huang, a Megaspeed executive, to gather information on US AI chips. Nvidia denies the claims and says the contacts were ordinary business meetings with no sensitive information shared.

Why it matters: Scrutiny of the main AI chip supplier could affect how Washington handles chip sales to China.

Sources: Axios, The Verge, The Information

The Information: TSMC weighs making 3D DRAM memory chips for Nvidia

The Information reports TSMC is considering making advanced 3D DRAM memory chips at Nvidia's request. The Verge explains that 3D DRAM stacks chips vertically to add bandwidth and capacity. The Verge notes the move would put TSMC against Micron, Samsung and SK Hynix during a global memory shortage.

Why it matters: More memory supply could eventually ease the high prices caused by the shortage.

Sources: The Verge, The Information, Bloomberg

Amazon lets shoppers text its Alexa shopping agent

Amazon is letting customers text requests or photos to Alexa for Shopping, which finds, compares and orders products for them. The feature starts with a small group of invited US customers, and Amazon says millions more will follow in coming weeks. GeekWire cites Digiday as reporting that sponsored products can appear among the agent's suggestions.

Why it matters: Shopping through a text thread gives Amazon a new channel for both sales and ads.

Sources: The Verge, GeekWire, The Information

Thanks for reading, and we will be back tomorrow with the next digest.

---

About ScrutifAI: ScrutifAI is an independent third-party agency that tests software for AI agents, building a single access point where agents and people can access and use tested tools without learning each one individually, at a fraction of the cost of buying them directly.