Scoop
AI tool news · rumor vs. reality

The AI Wire ●

Rumors tracked. Announcements verified. Updated daily.

2026-11-14

ANNOUNCED

Italy fines Meta €805 million ($887M): the Facebook app read Android users' photo galleries without specific consent

Italy's data protection authority, the Garante, announced on Saturday, November 14 that it has fined Meta Platforms Ireland €805 million — about $887 million, and the largest fine the regulator has ever imposed. The Garante found that the Facebook app on Android accessed users' photographs, videos and their metadata through a suggestions feature without obtaining specific consent or properly informing them.

The feature, 'Get suggestions from your gallery when you browse Facebook', has been switched on by default since 2023 and reached about 20 million Android users in Italy. The case began with a July 2025 complaint from a consumer group in Bolzano, and the proceedings ran about 16 months. The Garante ruled that a single app-wide permission, a vague notice and a default-on setting did not add up to valid consent under Article 122 of the Italian Privacy Code.

Meta has 60 days to keep the feature off by default and to seek separate, specific consent, and it must report back after one, two and three months; Meta says it will appeal. The fine is about 0.47% of Meta's worldwide turnover. One limit matters: the investigation did not examine whether gallery content was used to train AI models — Meta denies that — and a newer gallery feature that processes images in the cloud will be assessed separately.

RELEASED

Artificial Analysis Coding Agent Index v2.1: Claude Fable 5.2 and Opus 5.5 tie for first at 58 on refreshed, harder tests

Artificial Analysis has released version 2.1 of its Coding Agent Index, which scores model-plus-harness combinations across roughly 290 tasks in three equally weighted benchmarks: FrontierCode, on real open-source repositories graded by whether maintainers would merge the change; Terminal-Bench, on agentic terminal work; and SWE-Atlas-QnA-AA, on answering questions about unfamiliar codebases.

The headline is a tie at the top: Claude Code running Claude Fable 5.2 and Claude Code running Claude Opus 5.5 share first place at 58, level with Devin Fusion CLI, which pairs Opus 5.5 with GPT-6 Luna. Opus 5.5 led version 2.0 at 61.7, but v2.1 refreshes the underlying benchmarks, so the lower top score reflects a changed test, not evidence that any model got worse — scores are not comparable across versions.

Price separates the leaders more than points do. Per the Daily Fledgling's summary of the index, Claude Code with Fable 5.2 costs $2.89 a task at medium effort and $44.22 at max effort, while Codex with GPT-6.1 Sol scores 52 at max effort for $1.51 a task.

RELEASED

Anthropic makes Claude Opus 5.5 the default model on paid plans — and tells users to drop 'think hard' prompts

Anthropic has made Claude Opus 5.5 the default model on its paid Claude plans — Pro, Max, Ultra and Team — according to The Decoder's November 14 report on the change and a new Anthropic prompting guide. Subscribers who never touch the model picker now land on the flagship Opus tier by default.

Alongside the switch, Anthropic advises users to remove phrases like "think hard" from their prompts and to start at Medium effort instead. The company says Opus 5.5 scales its reasoning to the task on its own, that in its tests Opus 5.5 at Medium effort matched Opus 5 at High, and that responses without the extra urging started faster with no measurable quality drop.

Anthropic also argues the economics can work in users' favour: a more capable model can finish a job in fewer steps, offsetting its higher per-token price, and Sonnet pushed to high effort can cost about as much as Opus 5.5. Those cost and quality figures are Anthropic's own test results, reported as the company's claims.

2026-11-13

ANNOUNCED

xAI cuts about 60 from the Grok Imagine video team as Musk shifts focus to coding tools and Grok Bot

Elon Musk's xAI has laid off roughly 60 staffers from its Grok Imagine video-generation team in the past week, The Information reported on November 13, leaving only a handful of employees to keep the product alive. Neither xAI nor SpaceX has formally announced the cuts or said what they mean for Imagine's roadmap.

The cuts come as the company reprioritizes toward its AI coding tools and the Grok Bot agent product, which shipped its team-focused 1.0 release on November 11. Reporting describes a broader reorganization of the combined company's AI operation into four focus areas: Grok main and voice, coding, Imagine, and Macrohard.

The reorganization has been running since xAI merged into SpaceX in February 2026, in a deal that valued the combined entity at about $1.25 trillion. With no public roadmap update after the cuts, which video capabilities continue — and on what timeline — is unanswered.

Update, November 14: users are feeling the cuts directly. Starting around November 12–13, many Grok users — paid subscribers included — reported the Imagine tab gone from the sidebar on web, Android and iPhone, with grok.com/imagine now opening the normal chat with Image and Video mode selected. Imagine still works inside chat and on the API, and xAI's status page listed it as operational on November 14, but xAI has not explained the change and its release notes do not mention it.

RELEASED

Sierra releases Poppy 1.0, a Personal Agent Protocol — Shopify turns it on for every merchant, and Meta's Muse looks for it first

Sierra released version 1.0 of the Personal Agent Protocol, known as Poppy, on November 13. Poppy is a shared standard for how a person's AI assistant deals with a company — a common language an agent can use instead of improvising on each site it visits.

The launch arrived with two large backers already switched on. Shopify has turned Poppy on for every merchant on its platform, so each store automatically publishes a poppy.json file describing what it sells and how an agent should work with it, and Meta's Muse agent now looks for Poppy first on every site it visits before trying other approaches.

One caveat is in the release itself: the specification is still labelled draft 1.0, and the files Shopify stores declare protocol version 0.4, not 1.0. The standard is not yet frozen, so details could still change before companies build on it for good.

ANNOUNCED

OpenAI disbands its stand-alone robotics team, betting GPT-6 Astra can run robots directly

OpenAI has disbanded its stand-alone robotics team and folded the work into the rest of the company, Fortune reported in a November 13 exclusive that OpenAI confirmed when asked. Robotics researchers move to the main research team under Chief Research Officer Mark Chen, while the applied robotics and hardware teams move to the Applied organization under President Greg Brockman.

OpenAI's stated reason is a model bet, not a retreat: a spokesperson told Fortune that researchers assumed early this year that robotics would build on video models like Sora, but general-purpose frontier models including GPT-6 Astra have since shown they can be applied to robots directly, so a separate team building robot-specific models is no longer the fastest structure. Robotics remains part of the long-term vision, the company said, and Sam Altman's target of a first public humanoid demo in 2027 stands.

The evidence Fortune cites is an Nvidia study in which an Astra-powered agent, given camera images, robot state and a control interface, reached 74.8% on a simulated manipulation benchmark (RoboLab-120), against 51.0% for the best specialized robot models. That was simulation, not a physical robot, and the study itself was not linked in the report. This is OpenAI's second robotics wind-down: the first team closed in 2020, a stand-alone team was rebuilt in 2024, and its Figure partnership ended in February 2025. OpenAI has published no blog post of its own and gave no headcount.

RELEASED

OpenAI ships Codex CLI 0.169.0: side-chat tabs that survive a restart, task search and a trust fix for codex exec

OpenAI shipped Codex CLI 0.169.0 on November 13, at 08:37 UTC — less than five hours after 0.168.0 the same morning. The opt-in terminal workspace now holds multiple side-chat tabs that can be switched by keyboard or mouse and reopened after a restart, and the agent command center can search older tasks by name, prompt preview or working directory.

The full-screen transcript can collapse completed turn details under each prompt while keeping answers visible and tool output searchable. Developers can also add or edit custom model providers from inside the terminal, including creating a provider during sign-in and using a device code for supported gateways over SSH or without a local browser.

The release also tightens a trust boundary: codex exec now keeps project configuration and MCP servers disabled in directories the user has not trusted, even with --sandbox workspace-write, unless the new --trust flag is passed. Trackers reading the release notes found no mention of GPT-6.2, which remains unannounced.

RELEASED

OpenAI ships Codex CLI 0.168.0: side-by-side chats, voice-note attachments and a model-provider picker

OpenAI released Codex CLI 0.168.0 early on Friday, November 13. The headline addition is a split-pane workspace: enable the tui_workspace feature to view a side conversation beside the main chat, with a draggable divider and a zoom shortcut for the focused pane.

The release also adds a model-provider picker — browse providers under Sessions, then Model providers, and choose the default for new chats, while open and resumed chats keep their saved provider — plus the ability to attach audio recordings when using an audio-capable model, and animated GIF generation and editing with frame-count and timing controls behind a feature flag.

It follows 0.167.0 on November 6, which added the agent command center. Trackers comparing the two versions found no subcommands or global options added or removed.

RELEASED

OpenAI launches ChatGPT Wallet: dots agents can pay for purchases with your cards — with a security check on every buy

OpenAI launched ChatGPT Wallet on November 13, a payment feature that lets its dots agents make purchases on a user's behalf without ever seeing the user's actual card details. It was built with American Express, Mastercard and Visa, and OpenAI calls it the first wallet that lets agents pay without access to real card numbers while preserving card rewards and dispute protections.

How it works: after the user reviews a purchase and passes a security check — Face ID or a fingerprint — the card network issues a single-use token with a scoped card number, expiry and security code for that transaction. At launch the agent cannot complete a Wallet purchase without that per-purchase review. OpenAI says delegated buying under pre-set spending limits, categories or time periods may come later, but it is not part of this rollout.

Wallet is rolling out starting today inside dots, initially to ChatGPT Pro subscribers aged 18 and older in the United States with eligible US-issued cards. OpenAI published a Commerce Partner Playbook for businesses alongside the launch, is working with fraud partners Forter, Riskified, Sardine and Signifyd and payment providers Adyen, Checkout.com and Stripe, and says it is taking no percentage cut on transactions at launch. OpenAI says 1.2 billion people use ChatGPT each week.

RUMOR Rumor still unresolved

Rumor: 'muse-spark-1.4-internal' surfaces in a Meta RocksDB pull request — Meta has announced nothing

Rumor: an unreleased Meta model identified as muse-spark-1.4-internal has surfaced in the test notes of a public GitHub pull request on Meta's RocksDB repository, first spotted by @LuminaBench and reported on November 13. The pull request, #15674, makes a small logging change to RocksDB's experimental live-tuning feature, adding the name of the model that answered and its usage to each recommendation.

The notes, which appear to be auto-generated, describe a real internal call: 33,535 input tokens, including 17,780 cached, producing 8,169 output tokens — 7,867 of them reasoning tokens — in about 65 seconds. That is evidence the 1.4 version is already running inside Meta's own tooling, not a benchmark: the PR contains no performance comparison with Muse Spark 1.3, which Meta released on September 2.

Meta has not confirmed a name, release date, pricing or feature set for Muse Spark 1.4, and the pull request was closed without merging a few hours after it opened, which is common for Meta changes that land through internal systems. Treat this as an internal-testing sighting, not a launch.

RELEASED

Meta quietly puts Muse in Chrome: the Muse extension is live on the Chrome Web Store, with no formal announcement

Meta's Muse personal agent is now available as a Chrome extension, 'Muse from Meta,' live on the Chrome Web Store — version 2.0.65, updated November 12 — even though Meta has not formally announced it. TestingCatalog spotted the listing on November 13 and noted Meta may still introduce it as part of a broader Muse update.

The extension opens Muse in Chrome's side panel so users can pick up existing conversations alongside the page they are on and hand off tasks such as filling out forms, comparing prices and following up on refunds. Meta's own help pages document the extension: when Muse browses through it, it uses the user's personal Chrome in its own tab group and can use sites the user is already logged into.

Meta's help page also sets the privacy bounds: Muse in Chrome can't see saved passwords, browsing history or Chrome settings, and passwords on a page are stripped from the text before Muse reads it. Muse was previously available through its mobile apps, desktop app and the web.

ANNOUNCED

LangChain raises $300M at a $3B valuation to turn LangSmith into the operating system for agents

LangChain announced on November 13 that it has raised a $300 million Series C at a $3 billion valuation, led by CapitalG, Alphabet's growth fund, with GIC, DFJ Growth, IVP and Sapphire Ventures participating and Salesforce Ventures, Insight Partners, QuantumLight and MongoDB Ventures among the new investors. Total funding to date is $460 million. The Wall Street Journal first reported the round on November 12; the company confirmed it the next day.

The valuation is more than double the $1.25 billion LangChain carried when it raised $125 million about a year ago. The money goes into LangSmith, its platform for running, evaluating and governing agents in production — agent runtime built on LangGraph and its open-source frameworks, plus tracing, version comparison and failure diagnosis for agents already in service. LangChain says it will show new LangSmith capabilities at an event on November 20.

The usage figures are the company's own and were not independently audited: LangChain says more than 7,000 organisations use LangSmith, including Nvidia, Toyota, Salesforce, Workday and Harvey, and that the platform handles over 1 billion events a day. No revenue or profitability figure was disclosed, so the $3 billion price is an investor bet on agent infrastructure becoming its own category, not a measurement of returns yet.

RELEASED

Google connects Google Wallet to the Gemini app: spending-informed answers for all US users, read-only and opt-in

Google announced on November 13 that all Gemini app users in the United States can now connect Google Wallet under Personal Intelligence, so Gemini's answers can be informed by their spending habits. Wallet joins Gmail, Photos, YouTube and Search on Gemini's connected-apps list.

Users tag @Google Wallet in a chat to ask about personal finance and budgeting. For deeper insights they can optionally link bank and credit-card accounts to Wallet through Plaid. The connection is opt-in in Connected Apps settings and can be turned off at any time.

Google stresses the access is read-only: Gemini can read information from the connected accounts but, in Google's words, can't move, manage or invest your money. The rollout covers web and the Gemini app on Android and iOS, for personal accounts; Google has not announced availability outside the US.

RELEASED

Google DeepMind opens the datasets behind WeatherNext 3: SatWeave satellite mosaics and PARDIG rainfall estimates are free to download

Google DeepMind released two datasets that feed its WeatherNext 3 weather model on November 13: historical archives and continuously updated live feeds for SatWeave and PARDIG. WeatherNext 3 itself launched on September 3; what is new is that the data underneath it is now openly available to researchers and developers.

SatWeave merges imagery from geostationary weather satellites — GOES-East, GOES-West, Meteosat and Himawari — into one global mosaic across 17 channels, with hourly live updates delivered about an hour behind real time and an archive running from 2016 to 2025. PARDIG estimates near-surface rainfall from satellite imagery and atmospheric data at 15-minute intervals and roughly 10-kilometre resolution, for regions with little ground-radar coverage.

DeepMind reports that PARDIG improves on NASA's IMERG-Final research product by more than 30% on hourly detection skill for light-to-moderate precipitation — the company's own figure, from a preprint that has not yet had third-party replication. One practical caveat: PARDIG's live feed updates only every six hours, with an average delay of about five hours, so fine time resolution is not the same as real-time delivery.

Both datasets are free from Google Cloud Storage, and PARDIG is also on Google Earth Engine, where its near-real-time archive covers January 2016 to October 2026 and a longer reanalysis version reaches back to 1999.

ANNOUNCED

Google's Antigravity agent teams help prove a 2013 conjecture and formalise the Weil conjectures — with the biggest claim still unpublished

Google Research reported on Friday, November 13 that teams of specialist Gemini 4 agents working in Antigravity, its AI coding environment, have helped researchers tackle long-open problems across mathematics, chemistry and software. The system, called /teamwork, splits a hard job between agents with separate roles — producers that draft solutions, critics that look for flaws, and auditors that check results from a clean context — using objective checkers such as the Lean 4 proof language to verify each step.

The headline result is the Courtade–Kumar conjecture, posed in 2013. Researchers from Google, CUHK, USC and Carnegie Mellon proved it in full generality, and the November post adds a machine-checked formalisation of the whole argument in Lean 4, which Google puts at 50 million lines of code. An independent check confirmed the lines are real but found only about 239,000 are human-readable proof; the rest is generated certificate data the Lean kernel checks mechanically, though the formalisation is public, rebuilt cleanly and accepted by the Lean Foundation's external checker.

Google also says a framework called /teamwork-LEAP, paired with Gemini 4 Argon, fully autonomously produced a machine-checked formalisation of the Weil conjectures — about 1.8 million lines of Lean across 28,800 subagent invocations. That claim is thinner: Google's post links to no repository, paper or statement file, so it cannot yet be independently checked, and a separate group, Axiom, published its own public Weil formalisation on November 6. Google says /teamwork and /boost are available in Antigravity now, and it has opened applications giving 10,000 academic researchers six months of free access to its highest tier.