Scoop
AI tool news · rumor vs. reality

The AI Wire ●

Rumors tracked. Announcements verified. Updated daily.

2026-11-03

ANNOUNCED

Microsoft retires the standalone Word, Excel and PowerPoint agents — document creation moves inside Copilot Chat

Microsoft is replacing the standalone Word, Excel and PowerPoint agents in Microsoft Copilot with document-creation and editing capabilities built directly into the Copilot app, according to admin message centre notice MC1536062 published November 3. Users with a Microsoft 365 Copilot licence will ask Copilot Chat to create or refine a document, workbook or presentation directly, instead of Copilot invoking a separate agent for each app.

The change affects every user with a Microsoft 365 Copilot user subscription licence who currently has access to the standalone agents. Microsoft says the update introduces no changes to data processing.

The rollout is staged: worldwide general availability begins in mid-November 2026 and is expected to complete by late December 2026. It is the latest step in Microsoft's consolidation of Copilot into a single app and a single chat surface, following the August merger of the consumer Copilot and Microsoft 365 Copilot apps.

ANNOUNCED

Meta raises $15 billion in a five-part bond sale to fund its AI build-out

Meta said on November 3 that it is raising $15 billion through a five-part US dollar bond offering, turning to debt markets to help finance its accelerating AI infrastructure spending, Reuters reported. The sale includes $3.25 billion of notes due in 2031 at a 5.85% coupon and another $3.25 billion due in 2033 at 6.25%, according to a regulatory filing.

It is Meta's third big trip to the bond market in little over a year, after a $25 billion sale in April and a $30 billion offering last year. The new debt also lands days after Meta expanded plans for what could become a $97.5 billion equity raise.

Meta has said it expects capital expenditure of $135 billion to $145 billion this year, largely driven by AI. Moody's rated the new debt Aa3, and Meta's total debt stood at $83.67 billion as of September 30, per LSEG data — a striking shift for a company that historically funded growth from cash on hand.

ANNOUNCED

Google flips the default for schools: Search, Gemini and Notebook off for K-12 accounts until an administrator turns them on

Google said on Tuesday, November 3 that Google Search, Gemini, Gemini Notebook and Gemini in Classroom will be switched off by default for K-12 student accounts, staying off until a school administrator explicitly turns each one on. The company said it was acting because schools, teachers, researchers and policymakers had asked for control, and published a 29-page report on AI in education the same day. Guided Learning, already available in Google Classroom from November 3, will also reach Search for student accounts through a pilot in selected US schools later in November. Google gave no date beyond 'the coming months'; The New York Times reported early next year.

The catch is in the bundle. Each product can be enabled separately, but the Search a school switches on still includes AI Mode and AI Overviews, which Google describes as part of the product — there is no way to give students classic Search and leave the AI layer off. 'Search and A.I. are deeply integrated,' Chris Phillips, Google's vice president for education, told the Times. New York City Mayor Zohran Mamdani accused Google of using monopoly power to force AI into public schools, and a Community Note on Google's own response post made the same point: schools cannot enable basic Search without the AI features.

The stakes are the habits of the education market: Gemini Notebook — the renamed NotebookLM — counts more than 30 million users, and defaults set in school accounts shape what a generation reaches for first. Districts that keep AI off entirely will now lose student access to Google Search altogether, and students will see a message saying it is not available.

Update, November 4: Google spent Tuesday playing defence. In a blog post titled 'Designing responsible AI for learning' and a post on X, the company said the policy was meant to give schools more choice, repeating that Gemini, Gemini Notebook and Search are off by default and that 'schools are in control,' able to turn each on as they see fit. It also said Search for schools is being updated for learning: AI Mode and AI Overviews will offer step-by-step Guided Learning through problems so students do not simply get quick answers to schoolwork. The response did not address the central criticism — that a school cannot switch Search back on without the AI layer that comes bundled with it.

ANNOUNCED

EU court adviser: Gemini's answers may breach publishers' copyright when they reuse articles in recognisable form

An adviser to Europe's highest court said on November 3 that Google's Gemini chatbot may breach publishers' copyright when it retrieves their articles and reuses them to generate answers, Reuters reported. Advocate General Maciej Szpunar's opinion in the case brought by Hungarian news operator Like Company draws a line: a publisher's reproduction right may be infringed when a chatbot gets around a paywall or other technical barrier, or reuses 'the form of expression of that publication in a recognisable manner.' The mere facts in a story are not protected — only their specific expression.

The same opinion cuts the other way on training: using copyrighted material to train large language models is 'freely permitted under the text and data mining exception,' Szpunar said, and there was no evidence Google had used Like Company's articles to train Gemini. The disputed activity covers June 2023 to May 2024. Google welcomed the training point and said Gemini 'learns linguistic patterns to create original outputs,' with publisher opt-out controls; the European Publishers Council argued publishers should not have to choose between search visibility and protecting their rights.

An advocate general's opinion is advisory, not a judgment — but the Court of Justice follows the majority of them, and it will rule in the coming months. A ruling against Google could force it to pay fees for content reused in AI answers, at a moment when publishers say AI summaries are replacing the search traffic they used to receive.

RELEASED

Google brings Dreambeans to India — its first market outside the US, in 14 languages

Google announced on November 3 that Dreambeans, its experimental Google Labs app that creates personalised daily story collections, is now available to everyone in India aged 18 and over. India is the app's first market outside the United States, and it is available on Android and iOS in 14 languages: English plus Hindi, Bengali, Telugu, Marathi, Tamil, Urdu, Gujarati, Kannada, Malayalam, Odia, Punjabi, Assamese and Nepali.

Dreambeans brews its stories from the Google apps a user chooses to connect, including Calendar, Gemini, Gmail, Photos, Search and YouTube. The resulting daily collection can cover places to explore, arts and crafts, or reminders of upcoming local events, and users can dive deeper into a story, bookmark it for later, or share it with others.

The expansion is also a test of appetite. Dreambeans first appeared in the US in June, and Sensor Tower estimates cited by TechCrunch put US downloads at nearly 55,000 since then, with some stories now able to include short animations made with Google's Veo video model. Those same estimates suggest downloads fell sharply in October, so India — where Google says interest was strong — will show whether the daily-story format travels.

RELEASED

Google Antigravity starts rolling out Remote Agents — coding sessions that run in Google's cloud, not on your laptop

Google's Antigravity changelog for November 3 (version 2.29.0) says Remote Agents are starting to roll out to Google AI Pro and Ultra subscribers signed in with personal accounts. A conversation in a cloud project runs in an isolated, Google-managed cloud virtual machine instead of on your own computer, and can be opened from any browser at antigravity.google.com/remote; the feature is not on the free plan, Google AI Plus, or Workspace and Google Cloud business accounts. Google expanded the documentation on November 6.

That is different from Antigravity's existing Remote Control mode, which drives a session on your own powered-on workstation. Remote Agents keep long-running coding sessions, test suites and pull-request work going when the laptop sleeps or the browser closes; developers can attach a GitHub repository to a project, review diffs and approve permission prompts from a mobile browser. The same release adds Gemini Live voice mode, a redesigned Automations page and a native Jupyter notebook viewer.

The timing lines up with the wind-down of Jules, Google's earlier async coding agent: Jules runs sessions until November 11, then goes read-only ahead of a December 31 shutdown, with Remote Agents as the in-house successor path.

ANNOUNCED

Google expands its free AI Educator Series to Brazil, Mexico, Spain, Australia and sub-Saharan Africa

Google announced on November 3 that its AI Educator Series, a no-cost on-demand training programme giving teachers practical AI skills, will expand in the coming months to Brazil, Mexico, Spain, Australia and sub-Saharan Africa. The series first launched in the US earlier this year and has since reached India, Korea, Malaysia, the Philippines, and the UK and Ireland.

Teachers already in the programme are using Gemini and Gemini Notebook to turn topics into comic strips, podcasts and infographics, build lesson plans and assessments, and cut preparation time. Google says September sessions in India trained 80,000 educators, Malaysian teachers report saving three to five hours of prep per topic, and a partnership in the Philippines aims to bring AI training to over one million educators.

New November modules go live in the US on November 4, focused on saving teachers time: building interactive assessments with instant feedback, guiding students through game-based learning, and spotting assignment questions that invite shortcuts. Educators who complete the training earn a badge and join the programme's global community.

ANNOUNCED

Election Day: data centre pauses are on the ballot in the governor's races — backers of a moratorium could win up to 15 states

AI data centres have become a campaign issue in today's US midterm elections. The Information's analysis of candidate platforms and public statements, published November 1, found at least one major-party nominee backing a data centre moratorium in 21 of the 36 states electing a governor today — in Michigan, New Hampshire, Ohio, Oregon and Texas, both nominees back a pause — and concluded moratorium backers could win in up to 15 states, with pause supporters polling ahead in Maine, New Mexico and New York and seven more races close.

The underlying numbers are contested, and worth reading with care. SiliconNoon notes the dataset both The Information and WIRED draw on — MultiState's candidate tracker, published September 18 — flags a moratorium candidate in 16 states and roughly 20 candidates in total, with only five or six of those races genuinely competitive. Of the candidates not calling for a moratorium, most back at least one other restriction, most commonly ratepayer protection so campus power costs do not land on household bills; Hawaii is the only state where neither nominee campaigns on data centre limits.

Three sitting governors have already acted: New York's Kathy Hochul, Texas's Greg Abbott and Oregon's Tina Kotek have all paused approvals for new large-scale data centres, and Anthropic listed state-level moratoriums as a business risk in its IPO prospectus last week. Whatever today's results, the count that matters for builders is local as much as statewide — Ohio alone has at least 125 local moratoriums in effect and 14 local ballot measures today.

ANNOUNCED

Apple opens Siri AI to every app developer in a 27.3 developer preview — customers get it 'in a few months'

Apple announced on November 3 that any developer can now plug their app into Siri AI through App Intents, in a Developer Preview shipping with the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas. Until now, Siri AI integrations were limited to a short list of launch partners Apple hand-picked; the preview removes that gate and needs no new entitlement.

The mechanism is a new Swift macro, @AppIntent(.tool), which turns a custom App Intent into something Siri AI can call directly, with personal context, onscreen awareness and natural language on top. New in this preview are 'skills' — a way to teach Siri AI to combine several of an app's actions into one experience for requests too complex for a single intent — and Spotlight semantic-index lookups that can surface any entity an app has indexed, not just ones fitting Apple's predefined schemas. Apple's own apps, including Calendar, Mail, Maps, Messages and Photos, already use the system.

The limits are substantial. Siri AI needs an Apple Intelligence device set to a supported language — English, French, Japanese, Korean, Portuguese or Spanish to start — and it is unavailable in the EU on iOS, iPadOS and watchOS and to users under 13. Server-side features carry daily usage limits, with paid expanded access promised later. Apple says the capabilities ship to customers 'in a few months,' and App Store submissions for Siri AI experiences do not open until spring 2027.

ANNOUNCED

Claude Code product lead Cat Wu leaves Anthropic — the third departure from the team in two weeks

Cat Wu, Anthropic's head of product for Claude Code and one of the tool's two public faces alongside its creator Boris Cherny, announced on X on November 3 that she has left the company after two years. 'Friday was my last day at Anthropic,' she wrote — Friday was October 30. Wu also oversaw Cowork, Anthropic's knowledge-work agent for non-engineers. Anthropic confirmed the departure to Business Insider; Wu gave no reason and named no next role, answering a question about taking time off with 'post agi!'

Wu joined in the summer of 2024, when Claude Code was still a small internal tool, and fronted its February 2025 research-preview launch video with Cherny. In her farewell post she charted how the product changed under her tenure: early users manually approved every permission prompt and mostly used Claude for questions and test-writing, while today auto-approve is the default and Claude handles large migrations and builds new products.

She is not leaving alone. Claude Code engineer Thariq Shihipar and desktop developer Anthony Morris both announced their own departures on October 24, so three members of the team have gone in under two weeks — weeks before Anthropic's planned IPO, with trading expected as early as the week of November 16. No successor has been named and no roadmap change has been announced alongside the exits.

Update, November 5: Wu has said where she went. At 00:25 UTC she posted that she is joining OpenAI to lead product management for Core Products, working with Thibault Sottiaux — Tibo — and OpenAI's product team to build agents that 'have your back and your team's, and keep you in control.' The post passed 2.5 million views within ten hours; OpenAI confirmed the hire to Business Insider, per later reporting, while Anthropic still has not named her successor. Austin Lau, who helped build Anthropic's growth function, also posted the same night that he had recently joined OpenAI.

2026-11-02

RELEASED

xAI quietly reroutes Grok Imagine's Quality model to Image 2.0 at low quality — requests succeed, the model answering is different

Since November 2, API requests naming xAI's grok-imagine-image-quality model no longer reach that model: they keep the same request and response shape but are served by grok-imagine-image-2.0 with quality set to low, and the older grok-imagine-image-pro name follows the same chain. Nothing returns an error, and the bill drops by a cent per image at every resolution — the replacement bills $0.04 against the old $0.05 at the standard size, AI Pricing Guru's daily-verified pricing tables show.

xAI did give notice: its migration guide went up with a 60-day window starting September 2 and recommends switching to the new model by name so customers choose the quality they pay for. Even so, this is the third xAI model retirement in five weeks handled by silent redirect rather than an error, one tracker noted, and xAI keeps no dedicated deprecations page and no written general notice policy.

The practical lesson for developers is to log the model that actually answered, not just the one requested — at xAI right now, the two can differ. The redirect lands in the same week xAI switches on a default text watermark for Grok API output, from November 7, which the Wire covered on October 31.

ANNOUNCED

xAI introduces Data Bot, a shared analytics bot inside Grok Bot built by its own data science team

xAI introduced Data Bot on November 2 as a Shared Bot inside Grok Bot, built by the SpaceXAI data science team to handle warehouse queries and analytical work for colleagues across the company. The team said its small size had become a bottleneck for straightforward questions that still needed specialist knowledge, so it encoded that knowledge into a bot anyone can message.

What Data Bot carries is a maintained, shared skill set: metric definitions and query logic, guidance on setting up A/B tests and checking results for bias, chart and dashboard conventions that can turn a chat answer into a Databricks or Hex dashboard, and dbt and data-engineering practices. A separate shared memory holds operational knowledge — who owns a pipeline, whether a past metric dip was a regional holiday — with team managers approving or rejecting proposed additions before they reach everyone.

xAI's examples are internal work, not a public product launch: a product manager investigating questions dashboards do not answer, a lifecycle marketer tracing what Braze campaign recipients did next during The X Subscription launch, and scheduled weekly metric reports posted into Slack channels. The company estimates Data Bot handled about 70 hours of work in the week before the announcement that would otherwise have gone to the data team, while complex analyses still stay with the humans.

ANNOUNCED

WIRED: 400+ ordinary-looking apps hid deepfake 'nudify' tools — Apple pulled 19, Google had 350,000 downloads' worth

A WIRED investigation published November 2, built on research by the nonprofit AI Forensics and cybersecurity firm Infoblox, found a network of more than 400 connected apps and websites posing as everyday tools — pet-photo editors, language tutors, workout planners — that could secretly load interfaces for making sexually abusive and violent deepfakes from uploaded photos. A server-side switch turned the hidden features on after the apps passed store review, in some cases triggered by arriving through a Meta ad.

The scale ran through both major stores. On Google Play, 27 of the apps were live in September with more than 350,000 combined downloads, including DiffAI at over 100,000 installs; Google had removed most of them before WIRED got in touch. Apple removed 19 apps after WIRED contacted it, including the pet-photo app Viefur, and said the apps complied at review and were changed afterwards — adding it has suspended tens of thousands of apps for similar bait-and-switch tactics in the past year.

Researchers traced the network to Chinese-language codebases and a Hong Kong-registered company, KeLing Technology Limited, registering as many as 30 domain-and-app combinations a week at its peak. Meta said it banned KeLing in 2025 after a cease-and-desist, and reported acting against 11,600 ad accounts, about 4,700 business pages and 209 apps on its advertising platform, while WIRED found more than 84,000 ads tied to the network had run on Meta since January 1.

ANNOUNCED

SemiAnalysis launches GDPx, a 206-task agent benchmark where the best model scores just 12% — Claude Opus 5.5 leads

SemiAnalysis released GDPx on November 2, a benchmark for general agentic capability spanning 11 domains the firm picked for where frontier models will earn their money: chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement, hardware design, materials science, alignment and software engineering. Version 1.0 has 206 tasks built with 14 specialist data partners, an eight-hour default time limit, and models running in their native harness — Claude Code for Claude, Codex for OpenAI's models, Antigravity for Gemini. New versions are planned every three to six months, with a public leaderboard at gdpx.semianalysis.com.

The headline number is how low the top score is: Claude Opus 5.5 leads at 12.0%, ahead of Gemini 4 Carbon at 10.9% and GPT-6 Astra at 9.6%. No model leads everywhere — Opus 5.5 wins quant trading, hardware and chip design, Carbon wins biology and edges software engineering, Astra wins recursive self-improvement and materials science — and under binary grading, finance and legal came back at essentially zero for every model. SemiAnalysis also notes Opus 5.5 and Astra are held back in cyber, biology and alignment by their own safety classifiers: strip those three domains out and Astra edges Opus 5.5, while Meta's Muse Wonder Full is rated the best available cyber model. Opus 5.5's lead is not cheap, at $51.41 per task against roughly $17 for Carbon.

Two caveats belong with the headline. GDPx is a paid SemiAnalysis product whose authors want it to become the benchmark of record, and its partner-built tasks are not independently audited. And a 12% top score reflects tasks deliberately chosen to be frontier-hard — it measures how far agents are from full automation of expert work, not whether they are useful today.

ANNOUNCED

OpenAI publishes LifeVal, a benchmark for everyday chores — GPT-6 Astra scores 80.6% on OpenAI's own test

OpenAI published LifeVal on November 2, a benchmark that grades AI on everyday responsibilities instead of exams or code: finding childcare, enrolling a child in the right kindergarten, ordering groceries on a budget, managing household finances and getting through government paperwork. The task list was built from the American Time Use Survey, the Multinational Time Use Study and a new survey of 3,995 people on where they most want an assistant's help, across caregiving, health, education, household administration, personal finances, shopping and government services.

The headline score flatters its maker. GPT-6 Astra averaged 80.6% reward, up from 49.3% for GPT-5.2 in December 2025 and 80.0% for GPT-5.6, with example tasks such as organising flood volunteers at 98.0% and diabetes meal prep at 90.4%. OpenAI says it has used LifeVal internally for about a year to track and drive improvement on exactly these tasks.

The limits are structural: OpenAI built the test, chose the tasks, ran the scoring and publishes scores for its own models only, with no independent verification and no opening for third-party models to be graded. An average reward score is also not a share of real-world jobs finished correctly — LifeVal is best read as OpenAI showing which everyday jobs it is optimising ChatGPT for, not as a neutral league table.