Google ties school Search access to AI tools

Today's digest leads with Google's plan to remove Search for K-12 schools that keep its AI tools switched off. It also covers OpenAI's Images 2.6, Marvell's deal for Lumentum, new research on model editing, and data center ballot measures.

AI lab releases

Google will cut Search for schools that refuse its AI tools

Google plans to switch off its AI tools for K-12, or school-age, student accounts next year unless administrators turn them on. Schools that keep AI off will also lose Google Search, because the company says the two are deeply integrated. Senator Bernie Sanders called the plan extortion, while Google says the change gives schools control.

Why it matters: Schools that rely on Google's classroom products must now weigh AI limits against keeping a basic research tool.

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

OpenAI's Images 2.6 brings animated GIFs to ChatGPT

OpenAI released Images 2.6, an image model that lets ChatGPT users make animated GIFs from a prompt or a photo. OpenAI's own figures say the model makes images 50% faster than before. Developers get two versions through the API, the interface software uses to connect to the model: Flare for speed and Sunburst for quality.

Why it matters: Casual users and marketers can now make short animations inside a chat window.

Sources: The Verge, The Decoder, OpenAI

OpenAI code releases add GPT-6 Rosalind to its model catalog

Two OpenAI software kit releases, versions 3.46.0 and 7.51.0, add a model called GPT-6 Rosalind to the catalog. Both also add residency options, which set where data is kept, for Australia, India and Japan. The release notes do not say what Rosalind does.

Why it matters: Developers can spot a new OpenAI model name here, though its abilities remain unknown.

Sources: OpenAI on GitHub

Acquisitions + deals

OpenAI to buy developer tools maker Pierre Computer Company

OpenAI said it will buy Pierre Computer Company, which makes open-source tools, meaning code anyone can use and change, for developers. Two Pierre libraries, Trees and Diffs, already run inside Codex, OpenAI's coding agent, which is AI software that carries out tasks. The deal still needs regulatory approval, and SiliconANGLE says the terms were not disclosed.

Why it matters: Developers who use Pierre's tools will soon depend on OpenAI for their upkeep.

Sources: OpenAI, SiliconANGLE, The Decoder, The Information

Marvell agrees to buy laser maker Lumentum in all-stock deal

Chipmaker Marvell agreed to buy Lumentum, which makes lasers and parts for networks that send data as light, in an all-stock deal. TechCrunch puts the enterprise value, which includes debt, at $115 billion and the equity value near $80 billion. The deal is expected to close in the first half of 2028 if shareholders and regulators agree.

Why it matters: Data center buyers could see more of the networking supply chain sit under one company.

Sources: SiliconANGLE, TechCrunch, The Information

CrowdStrike to buy two AI agent security startups

CrowdStrike said it will buy Fiddler AI and Dreadnode, two startups focused on securing AI agents. The Wall Street Journal reports CrowdStrike will pay about $400 million for Fiddler and about $95 million for Dreadnode. Fiddler monitors what agents do, and Dreadnode runs simulated attacks to find their weak points.

Why it matters: Companies deploying agents may soon find agent monitoring bundled into the security tools they already buy.

Sources: TechCrunch, SiliconANGLE

Newcomer reports investors see Anthropic and OpenAI on different paths

Newcomer reports that investors backing both OpenAI and Anthropic see the two companies on different paths. Newcomer reports one such investor expects Anthropic's IPO, its first public share sale, before Thanksgiving. Newcomer says OpenAI, by contrast, is looking for a comeback story.

Why it matters: Anyone watching AI markets could soon see a public price put on Anthropic, if the report holds.

Sources: Newcomer

More funding

• SiliconANGLE reports Phaze raised $8.75 million in early-stage seed funding from Andreessen Horowitz to give AI agents and people remote access to powerful workstations. Sources: SiliconANGLE
• GeekWire reports Google is giving the Allen Institute for AI about $5 million plus $1.8 million in cloud credits to build an open-source climate model. Sources: GeekWire
• GeekWire reports Finpilot, which uses AI agents to handle paperwork for institutional investors, raised $10 million in a Series A round led by Lobby Capital. Sources: GeekWire
• SiliconANGLE reports Sentra, which limits AI agents' access to company data, raised $150 million in a Series C led by Capital One Ventures. Sources: SiliconANGLE
• SiliconANGLE reports Traversal, which builds AI agents to keep software infrastructure running, raised $120 million at a $1 billion valuation. Sources: SiliconANGLE
• SiliconANGLE reports Applied Compute, a cloud platform for training and running AI models, raised $350 million at a $3.25 billion valuation. Sources: SiliconANGLE

MCP + agent tooling

GitHub Copilot comes to Xcode's built-in coding assistant

GitHub says its Copilot coding assistant now works inside the built-in assistant of Xcode, Apple's developer tool. The integration is in public preview, starting with the Xcode 27.3 beta, and requires macOS Tahoe 26.6 or later. All Copilot plans are supported, and the separate Copilot extension for Xcode remains available.

Why it matters: Apple platform developers get a second way to use Copilot, inside Xcode's own assistant.

Sources: The GitHub Blog

Cloudflare adds tools to track and control personal AI agents

Cloudflare launched analytics that show site owners which personal agents, AI tools acting for a person, visit their sites. Owners can filter by agent to see the pages it requests and the errors it hits. Cloudflare is also adding finer access rules for supported personal agents, so sites can choose which to allow, limit or block.

Why it matters: Site owners can now weigh agent traffic case by case rather than blocking it all.

Sources: Cloudflare Blog

LangSmith's bring-your-own-cloud option expands to Azure and Google Cloud

LangChain says its LangSmith agent platform can now run in a customer's own cloud account on AWS, Azure and Google Cloud. The AWS version is generally available, while Azure and Google Cloud are in beta, an early test phase. Customer data such as prompts and datasets stays in their account, while LangChain handles scaling and updates.

Why it matters: Teams with strict compliance rules can keep data in house without running the software themselves.

Sources: langchain.com

Research

MIT researchers say GPT-6 Astra can edit other models directly

MIT researchers report that OpenAI's GPT-6 Astra can read and edit weights, the numbers that define an AI model, without retraining it. The researchers say changing 0.016% of a small language model's weights raised its math benchmark score from 7.9% to 48.1%. Astra produced no measurable gain on another small model, Gemma 3 270M.

Why it matters: Fixing a model could one day cost far less than retraining it, if the results hold.

Sources: Devsplainers (YouTube), The Decoder

Anthropic shares prototype to prove which model wrote an answer

Anthropic released FlashProve, a research prototype that uses cryptography to show which AI model produced a given answer. Labs would publish a fingerprint of each model, which does not reveal its weights. There are no deployment plans yet, at Anthropic or any other lab.

Why it matters: Users could eventually check that a paid service really ran the model it promised.

Sources: TechCrunch, The Decoder

OpenAI says its pipeline produced 948 Linux exploits in one run

OpenAI says an automated system using an upcoming cyber model produced 948 working Linux exploits in one run. OpenAI says these exploits could let an ordinary local user gain full control of a machine. OpenAI says the run cost about $1.2 million, and that human researchers reviewed the results before reporting flaws with patches.

Why it matters: Linux maintainers may face more bug reports as AI makes finding flaws easier to scale.

Sources: OpenAI

Other notable news

Bloomberg reports Apple is cutting internal AI budgets for staff

Bloomberg reports Apple reduced what its engineers can spend on AI each day to $200 from $300 in October. Bloomberg's sources say further reductions are planned, taking some allowances down to $100 daily. For staff in other departments, Bloomberg reports, Apple is considering a daily limit that could go down to $25.

Why it matters: Even large companies appear to be rethinking how much staff AI use is worth paying for.

Sources: TechCrunch, Bloomberg

Smartsheet unveils vsActual, a standalone AI product for executives

Smartsheet unveiled vsActual, an AI product that compares a company's plans with work tracked in tools such as Jira and Workday. It can be used without Smartsheet's main software and is aimed at senior executives. Smartsheet is also adding usage-based pricing, with a base fee plus credits, while keeping per-user plans.

Why it matters: Executives may eventually spot slipping projects without moving work onto a new platform.

Sources: SiliconANGLE, GeekWire

Voters back data center limits in midterm ballot measures

According to a tracker compiled by Redbud Consulting, 33 of 45 local ballot measures on data centers will restrict development. An NBC News poll found 76% of Election Day voters opposed data centers in their communities. Winners include Texas Governor-elect Gina Hinojosa, who backs a statewide moratorium, a temporary ban on building.

Why it matters: AI companies may find new data centers harder to build in many communities.

Sources: The Verge, Fast Company

Nothing says its first AI-native product ships in December

Nothing CEO Carl Pei says the company's first AI-native product will ship in December. Pei says it lacks a new screen and could not exist as an app. He calls the industry's search for an iPhone of AI a dead end.

Why it matters: Buyers should wait for details, since Nothing has not said what the product does.

Sources: TechCrunch, The Verge

That is all for today's digest, and we will return with more AI and tech news in the next edition.

---

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.