The AI Wire ●

Meta confirmed on Thursday that its Muse personal AI agent now connects to Plaid's data network, gaining direct access to more than 12,000 financial institutions across the US, Canada, and Europe — every major retail bank and a long list of fintech apps.

The integration lets users link bank accounts, brokerage balances, and card transactions directly to Muse, which can then answer questions about spending, run budgets, and surface anomalies. Meta framed the launch as opt-in, with users authorizing each connection through Plaid's OAuth flow — and for now, Muse only reads; it does not execute trades or move money.

It's the first time a Meta AI product has sat inside the consumer financial stack at scale, and a shortcut past the slow build: hundreds of millions of Instagram and WhatsApp users are within reach on day one. Google's Gemini has been inching this direction through Google Wallet, and OpenAI has been testing similar hooks with Stripe.

Meta has opened early-access signups for upcoming Muse features, days after announcing a wave of them at Meta Connect 2026. Users can ask Muse — "Can you let the Muse team know I want to be part of the Muse early access program?" — to get on the list.

The pipeline includes a video-chat avatar for Muse, many more shopping partnerships and connectors, computer-use control for the Mac app, Muse on Meta's AI glasses via a wake word, and a dedicated Muse email address. Instead of a limited beta, Meta is explicitly recruiting AI enthusiasts to try features first.

It's a land-grab tactic: the crowd most likely to try new features first also uses multiple AI apps, and Meta wants them inside Muse before rivals close the feature gap.

Coinbase CEO Brian Armstrong shared a dataset on September 25 showing Grok as the dominant harness for agentic trading on Coinbase, the largest US crypto exchange. Grok accounts for 59.5% of the agentic traders using Coinbase for Agents, the June-launched feature that lets AI agents trade, pay, and run workflows inside preset limits. Musk acknowledged the lead with a one-word post: "Cool."

Custom models trail at 21.7%, followed by Anthropic's Claude at 13.4%, Perplexity at 3.5%, and OpenAI's ChatGPT at just 1.5%. The figures are a striking early datapoint in the agentic-commerce race, with Grok far ahead of every rival lab. The numbers come from Armstrong's public dataset as reported by TheStreet.

SpaceXAI has launched a 'Grok Bot Template Rewards' pilot that pays creators whose public Grok Bot templates drive user activity. Under the program terms, effective September 25, X pays eligible creators in USD every two weeks based on how many people use their Bots and how often.

The program is invitation-only during the pilot, and payouts are discretionary — not a revenue share and not guaranteed income. Eligible creators must be 18+, hold an X Premium tier, and have an active X Money account, among other requirements.

X product evangelist Sawyer Merritt disclosed he earned $500 in two weeks from his own Home Robots template. It's xAI's latest bid to build an ecosystem around Grok Bot, which Musk claims is growing 100% month over month.

Google has released a dedicated Gemini app for Windows, giving its AI chatbot a permanent spot on the desktop instead of living only in a browser tab or phone screen. The app installs directly on Windows machines and can be summoned with a keyboard shortcut, mirroring how OpenAI's ChatGPT desktop app and Microsoft's Copilot work.

Once open, it functions largely like the web version: questions, text and image generation, document analysis, and help drafting emails or spreadsheets. The headline difference is accessibility — one keystroke instead of a browser tab — extending Gemini from Android, iOS, and Workspace into the desktop layer itself.

Free access covers core features; Gemini Advanced, bundled with a Google One AI Premium subscription, unlocks more capable models and higher usage limits. The launch continues the consolidation trend: OpenAI, Microsoft, Google, and Anthropic are all racing to be the default AI assistant already sitting on your machine.

Google is reportedly testing a new 'Gemini Tasks' mode that could handle work across desktop apps, with Computer Use permissions and Gemini Live built in. The finding comes from Android Authority, surfaced in news roundups on September 25.

A desktop Tasks mode would put Gemini in direct contention with the agentic computer-use tools Anthropic, OpenAI, and Meta are all pushing this year. Until Google acknowledges the testing, it stays rumor.

Resolution (October 13): confirmed in shipped form. Google launched a tasks mode in the Gemini app for AI Pro and Ultra subscribers alongside Gemini 4 Argon, letting Gemini keep working while the user is away across more than 40 connected apps. The shipped version is an in-app toggle rather than the desktop Computer Use mode described in September reports, so the feature landed, in a different form than first rumored.

Google has introduced Live Avatar for Gemini 3.8 Live, adding a real-time visual presence to its conversational AI. The feature combines Gemini 3.8 Live's native live-dialogue capabilities with low-latency streaming video, and it is pitched at enterprise use — a face for customer interactions and meetings. It follows last week's launch of Gemini 3.8 Live voice models.

The timing is hard to miss: Meta unveiled a Realtime Avatar for Muse at Connect the day before, and Google is answering immediately. The two labs are now racing to give their assistants animated faces as voice agents become the default interface.

On September 25, a long-term ChatGPT Pro 20x subscriber posted on OpenAI's own community forum that hard usage limits had appeared on the plan without advance notice. The subscriber said the Pro model option became unavailable, with ChatGPT showing a "Limit reached. Try again after 4:59 PM tomorrow" message.

The post says the documentation now refers to hard daily and weekly caps, with no visible usage meter before the limit hit and no email or in-product notice to existing subscribers. At launch, ChatGPT Pro was sold at $200 a month with "unlimited usage," the poster notes, and describes the caps as a product regression rather than a plan change. OpenAI has not publicly responded to the report.

UPDATE Sept 29: Confirmed in substance. OpenAI product lead Tibo Sottiaux announced the reopening of $200 Pro sign-ups with an official new usage calculation — tacit confirmation that usage math on the 20x plan has changed. Open questions (30-day notice, grandfathering) remain unanswered, and the original subscriber report was never contradicted.

Anthropic said Friday that Claude can calculate particle-scattering amplitudes through nine loops, announcing the result in a new post on its Science Blog. Loop order measures how far successive corrections in a quantum-field-theory calculation extend — and nine is deep into territory that is notoriously hard even for specialists.

The announcement is a capability claim aimed squarely at scientific computing: Anthropic is positioning Claude as a serious research tool, not just a coding and chat assistant. The company frames scattering amplitudes as the formulas theoretical physicists use to predict particle behavior, built up loop by loop.

Caveats apply: the number alone doesn't show what Claude calculated, how much human guidance was involved, or how the result was verified — reproducing a known expression differs from deriving a new result independently. Still, as a signal of frontier-science capability, nine loops is hard to ignore.

Adobe has expanded its Anthropic integration and launched a new connector for Google's Gemini, letting users control Creative Cloud apps through text prompts inside third-party AI assistants.

The updated Claude integration, announced Thursday, now offers 80 "tools" across Acrobat, Photoshop, Premiere, Express, and Lightroom — up from 50 at its April launch. Users can retouch images, resize videos for social, and convert PDFs without leaving Claude, with new interactive editor controls in Acrobat.

The new Gemini connector provides similar access for Google's AI assistant users, joining Adobe's existing connectors for ChatGPT and Salesforce's Slackbot. It's a bet that more design and document work will happen inside AI assistants rather than native apps — putting Adobe's AI connector footprint inside all three flagship-app ecosystems.

Leaks began circulating Thursday, September 24, when TestingCatalog found references to a new "Pro Max" tier buried in OpenAI's backend. Independent researcher Tibor Blaho spotted the same strings, and the AI leaks account Chetaslua posted screenshots of the tier inside ChatGPT's own subscription interface. OpenAI has not confirmed any of it.

The headline feature is "Fastest Work and Codex" — raw speed for OpenAI's long-running agent products, Work and Codex, rather than more usage. The tier also advertises maximum memory, 100GB of file storage, early feature access, and priority availability.

The rumor gained weight this weekend when paying users noticed ChatGPT's paid page had been quietly redesigned: the existing $100 and $200 tiers were renamed "Pro Standard" and "Pro More," and the old "5x/20x the usage of Plus" claims were replaced with a vague "more usage than Plus" — suggesting a tier reshuffle ahead of Tuesday's DevDay. Still no official word from OpenAI.

Resolved September 29: confirmed. OpenAI launched the $500 Pro 500 plan at DevDay with its highest usage allowance — 25x the ChatGPT Plus allowance — plus Ultrafast access. This is the tier the leaks described (early code strings even called it 'Pro Max').

Update (2026-10-08): Confirmed at DevDay 2026 (Oct 6) as the $500/month Pro 500 plan — 25x the Plus usage allowance and the only plan with Astra Ultrafast in Codex. The leaked "Pro Max" name was wrong; the leaked price and the "sells speed" positioning were right.

Fortune reported Thursday, citing multiple sources familiar with the plans, that OpenAI is set to preview GPT-6 Cyber, a cybersecurity-focused model, within days — with its DevDay conference in San Francisco on Tuesday, September 29, as the likely stage. Reuters carried the report on September 24. A limited group of customers in the application-only Daybreak Red program already has access for alpha testing.

GPT-6 Cyber would be OpenAI's fourth security model of the year, following GPT-5.4 Cyber in April, GPT-5.5 Cyber in June, and GPT-5.6 Cyber in August. Alongside it, OpenAI plans a first-of-its-kind enterprise security platform — a gateway environment for building automated defenses and patching vulnerabilities, modeled on how ChatGPT serves as the consumer gateway to frontier models.

Coverage has intensified ahead of DevDay: TechGig and Forkast now report the Sept 29 preview as the working plan, with chief revenue officer Dali Rajic leading a commercial push behind the security line. One caution from The Hack Academy: OpenAI's official DevDay page confirms the event and a livestreamed keynote but says nothing about GPT-6 Cyber — so the preview date remains unconfirmed by OpenAI itself.

Resolved September 29: confirmed in substance. No 'GPT-6 Cyber' model was previewed, but the enterprise-security side landed hard: Codex Security Cloud (scheduled repo scanning with dedup and cloud-prepared fixes, Daybreak defensive models included) plus Private Intelligence — zero data retention with Private Safety Processing — and a Private Inference preview coming this fall.

Update (2026-10-08): Still unresolved: GPT-6 Cyber did not appear at DevDay 2026 (Oct 6). Fortune's later reporting said it is expected "within weeks, not days." DevDay's security announcement was Codex Security Cloud instead.

Update: DevDay 2026 (September 29) came and went with no GPT-6 Cyber preview and no enterprise security platform. OpenAI's developer event centered on Dots, Codex updates, the Pro 500 tier and the Decisions API. Marking this rumor debunked.

Mark Zuckerberg announced on social media Wednesday that Meta is launching Muse Code, its first AI programming agent, in a direct challenge to OpenAI's Codex and Anthropic's Claude Code. The tool is in beta and is powered by Muse Spark 1.2, the latest base model from Meta Superintelligence Labs.

Meta's published benchmarks position Muse Code as a value-and-reliability play rather than a capability leader. On Terminal-Bench 2.1, Muse Code with Muse Spark 1.2 scored 82.9% — ahead of Codex on GPT-5.6 Terra (81.8%) and Grok Build (81.6%), but behind Claude Code on Opus 5 (86.7%). Meta's pitch is low price plus 'crash recovery' capabilities instead of top benchmark scores.

The move deepens Meta's expansion into the developer layer — its first standalone coding application — and opens a new monetization lane as it burns heavily on AI infrastructure. It lands a week after Muse topped the iPhone App Store, with Meta Connect framing Muse as a personal agent that now spans phone, smart glasses, and the developer desktop.

Google is testing 'Call for Me,' a feature that lets Gemini call businesses on a user's behalf — checking whether a store has a product in stock, making restaurant reservations, moving scheduled appointments, or placing items on hold. The rollout is initially limited to Pixel 11 owners in the US who pay for a Gemini subscription and use the beta version of Google's Phone app for Android.

The AI introduces itself to the business, navigates automated phone menus, waits on hold, and handles the conversation, while users watch a live transcript and can take over at any time. Notably, Gemini can share user-approved personal information during calls and dials from the user's own phone number — a broader mandate than earlier Duplex experiments.

The feature answers Meta's Muse, which made calling businesses a centerpiece at Connect, and comes as the big daily-use assistants converge on agentic calling. It builds on Google's years of experiments with AI-assisted calling, including the Duplex salon-booking demo at I/O and later Hold for Me features.

Spotify has plugged into Meta's Muse AI agent, making it the first major music streaming service with a Muse connector, following Ticketmaster as the platform's first music-related integration. Users who link their Spotify account through Muse can ask the agent to play music, save songs to their library, find podcasts or audiobooks, and build or edit playlists — all in conversation.

The more interesting part is scheduling. Spotify says Muse can tie listening to a user's calendar: a road-trip entry could spin up a destination-themed playlist that starts when the trip begins, and a 'Personal Podcast' feature can turn a user's notes and briefings into audio episodes stored in their library.

It's also a data-sharing story: linking Spotify hands Muse the keys to a user's music library, and some users have already called for an opt-out option. For Spotify, it's the third AI-assistant integration after its ChatGPT app and Anthropic Claude connector — the company clearly intends to be wherever people delegate to chatbots.