Last checked October 23, 2026 · News roundup
Today in brief: OpenAI's models and Codex are coming to Google Cloud, and Walmart opened a door that Amazon is closing: a gateway that lets approved AI agents shop for you. On the business side, Microsoft tied the ConnectWise Cloud Backup outage to a China-linked group running AI attack agents, OpenAI is reportedly raising at a $1.25 trillion valuation, and a judge ordered Meta to pay New Mexico $25 billion.
Every figure below was checked against the vendor post or the outlet named next to it on October 22–23, 2026. Where only one outlet has reported something, we say so.
Money raised or reportedly being raised, Oct 21–22, 2026
AI · 4 stories
Can you use OpenAI models on Google Cloud? Yes, in a private preview
OpenAI said on October 21 that its frontier models and Codex agent workflows are coming to Model Garden, Google Cloud's catalog of AI models. Companies buy them through Google Cloud Marketplace, and enterprises with Google Cloud spending commitments can count these purchases against those commitments. OpenAI says general availability is planned "in the coming weeks"; until then access goes through a request form, and OpenAI models will also come to Google Antigravity and the Gemini Agent.
OpenAI's post does not name the models or list prices. According to The Decoder, the launch covers GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna as a private preview for select enterprise customers, billed on a single Google Cloud invoice. Microsoft remains OpenAI's primary cloud partner, OpenAI notes.
Who it affects: teams standardized on Google Cloud that wanted GPT models without a second vendor contract. What to do: if you have unused Google Cloud commitments, ask your account team whether OpenAI usage will draw them down before signing a separate OpenAI deal. If you are weighing Google's own model instead, our breakdown of Gemini 4 Argon's price and who can access it has the per-token numbers.
Can ChatGPT or Gemini shop at Walmart for you? Agent Profiles arrive in November
Walmart CEO John Furner unveiled Agent Gateway, an MCP server that lets approved AI agents search Walmart's online inventory and buy for a customer, TechCrunch reported on October 21. Each agent gets its own Agent Profile inside your Walmart account the first time you use it, so its orders show up separately from yours, and you can delete the profile at any time.
You choose the control level: approve every purchase, or pre-authorize buys under a spending limit you set. Agents cannot see your payment details, may only touch the data a task needs, and get blocked if they break Walmart's agent terms. Walmart's normal returns policy still applies when an agent orders the wrong thing. Agent Profiles roll out next month, and a Walmart+ membership is not required. Walmart already works with Meta's Muse and has integrations with ChatGPT and Gemini, while Amazon has started blocking AI agents from buying on shoppers' behalf.
What to do: start with per-purchase approval and a low cap until you trust the agent. For ChatGPT's other shopping tool, see how ChatGPT Try On previews clothes on you.
Walmart Agent Gateway: what an approved agent gets
Allowed
✓ Search Walmart's online inventory
✓ Buy for you, in its own Agent Profile
✓ Spend up to a limit you pre-authorize
✓ Use the data a task needs, nothing more
Not allowed
✗ See your payment information
✗ Act beyond what you asked for
✗ Keep working if it breaks the agent terms
✗ Override Walmart's returns policy
What is Recursive Flint? A free coding model that runs on a laptop
Flint is the first model from Recursive, the startup led by former Salesforce chief scientist Richard Socher, released October 21 for free on Hugging Face under the Apache 2.0 license. It is a mixture-of-experts model with 40 billion parameters in total but only about 2.5 billion active per token, which is why it runs on a laptop. Recursive says its automated research system, the Eureka Machine, did most of the design work, including low-bit training that shrank the model by 71%.
In Recursive's own tests, Flint generated about 131 tokens per second on a MacBook Pro with an M4 Pro chip and over 150 on Nvidia's RTX Spark. On SWE-Bench Verified it beat other models under 50 billion parameters. The company admits Flint needs more guidance than paid cloud assistants and suggests splitting big jobs into small steps.
Who it suits: developers who want an offline coding model with no per-token bill. If you mostly code in an editor, compare it with what a paid tool like the Cursor AI code editor gives you before switching.
Local speed on a MacBook Pro (M4 Pro), tokens per second
DeepSeek funding: a reported $15 billion raise two days after the federal ban
DeepSeek has raised about $15 billion (roughly 100 billion yuan), Bloomberg reported on October 22, with battery maker CATL and Tencent among the potential investors, according to TechCrunch's summary. No valuation was given. For scale, Reuters reported in June that DeepSeek raised more than $7 billion at a $50 billion valuation.
The timing matters for US users: the round came two days after Washington ordered federal civilian agencies to remove DeepSeek technology from their systems, which we covered in our October 20 roundup. The money does not change that order. If your company follows federal procurement rules or has a policy against the DeepSeek app and API, nothing here reopens the door; open-weight DeepSeek models run on your own hardware are a separate question for your security team.
SaaS & business · 5 stories
Is ConnectWise Cloud Backup down? Microsoft ties the outage to Storm-0940
Yes, partly. ConnectWise switched off Cloud Backup (formerly SkyKick), SaaS Security and Cloud Management after attackers took over the apps' identities in Microsoft 365, and it says it is restoring them in phases over several days. On October 21 Microsoft attributed the attack to Storm-0940, a China state-aligned group, which used ARTEX, an open-source framework that runs AI agents for scanning, password spraying and app abuse.
Using the stolen app access, the group tried to sign in to more than 2,250 Microsoft 365 tenants at up to 2,900 attempts a minute, and in some cases went from first contact to privileged cloud access in under three hours. Microsoft says this was not a flaw in its own products, and that an affected tenant does not automatically mean data was taken.
What MSPs and IT admins should do now: keep the ConnectWise connections disabled until ConnectWise gives restoration steps; do not delete the enterprise apps first, because they are evidence; search tenants by application ID; export Entra sign-in and audit logs before they expire (7 days on free Entra, 30 on P1/P2, per ConnectWise's guidance as reported by Techzine); and check for new app credentials, consent grants, role changes and mail forwarding rules since October 1.
ConnectWise / Storm-0940 timeline
- September 2026 · Earliest ARTEX-driven activity Microsoft has found
- Oct 10–14 · Password spray: 327 accounts at 8 organizations, 151,000+ failed logons, no successful sign-in
- Oct 11 · Peak rate: 2,900 failed attempts per minute
- Oct 15 · Public reports of the ConnectWise compromise
- Oct 16 · ConnectWise publishes its first advisory
- Oct 17 · ConnectWise disables the affected app identities (per Techzine)
- Oct 21 · Microsoft names Storm-0940 and publishes mitigations
OpenAI valuation: a reported $1.25 trillion before new money
OpenAI is negotiating up to $30 billion in new private funding at a $1.25 trillion pre-money valuation, CNBC reported on October 22, with Coatue expected to lead and investors from the UAE taking part. Terms could still change, and OpenAI declined to comment. A month earlier Bloomberg reported a $1.4 trillion target, and OpenAI's March round valued it at $852 billion after the money came in.
If it closes above $1 trillion, OpenAI would be the most valuable private company ever. It filed confidential IPO paperwork in June but is raising privately first, TechCrunch notes. Rival Anthropic raised $65 billion at a $965 billion valuation in May and is reportedly seeking $1.5 to $2 trillion in its IPO. For ChatGPT users and API customers nothing changes today; prices and plans are set separately from funding.
Meta's $25 billion New Mexico penalty: what the ruling says
A New Mexico judge on October 21 ordered Meta to pay the state $25 billion, the largest payout yet tied to the Cambridge Analytica scandal. A jury had found after a two-week trial that Meta deceived New Mexico users about its privacy practices, counting 43 million violations of state consumer protection law at up to $5,000 each, so the award is well below the theoretical maximum.
Meta will appeal. Chief legal officer C.J. Mahoney said the ruling makes it "all but impossible to do business" in the state, while Attorney General Raúl Torrez said New Mexico "may have paid for the construction of schools for the next 100 years." Nothing changes in Facebook or Instagram for users or advertisers yet. The bigger signal for SaaS companies is that state consumer protection laws, applied per user, can produce penalties on the scale of federal ones.
Legora raises $550 million at $8.8 billion in the legal AI race with Harvey
Legal AI startup Legora raised a $550 million Series E at an $8.8 billion valuation, led by Atreides Management with Baron Capital, Bessemer, General Catalyst and Menlo joining, TechCrunch reported on October 21. Annual recurring revenue is now above $200 million, double the $100 million it reported in April, when an extension valued it at $5.6 billion.
The money goes into expanding across North America, where Harvey is the incumbent: Harvey raised $550 million at $15.5 billion about a month ago. For US law firms and in-house teams, two well-funded rivals usually mean sharper pricing and pilots; Legora has not published prices, so expect quotes per seat or per firm. Investors are still paying up across AI: on the same day, year-old research lab Flapping Airplanes raised $1 billion at a $6 billion valuation.
Automattic layoffs: what WordPress.com and VIP customers should know
Automattic, the company behind WordPress.com, WooCommerce and Tumblr, has started layoffs after a board shake-up and senior departures, TechCrunch reported on October 21. Cuts hit communications, social, WordPress.com and the VIP enterprise team, and more are expected over the coming months ahead of a debt renegotiation in June, according to insiders. CEO Matt Mullenweg said the number of terminations was not above normal; no headcount figure has been given. The chief financial officer, an interim CFO and the chief legal officer have left, and the board was fully replaced after a failed September attempt to remove Mullenweg.
No change to WordPress.com plans, prices or support has been announced. If your site runs there, this is a good week to confirm your export and backup work, and VIP customers should ask their account manager who now owns their contract.
Short answers to today's top searches
Can I use GPT-6 on Google Cloud today? Only if you are accepted into the private preview. OpenAI says general availability is weeks away and is taking requests through a form; The Decoder lists GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in the preview. Pricing has not been published, and purchases go through Google Cloud Marketplace.
Is it safe to let an AI agent buy things at Walmart? Walmart limits the risk: the agent works in its own Agent Profile, never sees your card, and can be held to per-purchase approval or a spending cap. Returns still follow Walmart's normal policy. Profiles arrive in November, so start with approvals on and a small limit.
Do I need to do anything if I use ConnectWise Cloud Backup? Yes. Keep the integrations off until ConnectWise sends restoration steps, export your Entra sign-in and audit logs now, and review new app credentials, consent grants and role changes since October 1. Microsoft says a connected tenant was not necessarily breached, but check rather than assume.
Is Flint really free? Yes. The weights are on Hugging Face under Apache 2.0, which allows commercial use and changes. Your only cost is the hardware: it runs on a recent laptop, and Wired's reviewer used a Mac with 48GB of memory. It needs more step-by-step prompting than paid cloud coding assistants.
More on agent tools and coding assistants lives in our AI agents and Claude Code hub.
Sources (accessed October 22–23, 2026): OpenAI, models on Google Cloud (Oct 21) · The Decoder (Oct 21) · TechCrunch, Walmart (Oct 21) · TechCrunch, Flint (Oct 21) · TechCrunch, DeepSeek (Oct 22) · Microsoft Security, Storm-0940 (Oct 21) · Techzine, ConnectWise (Oct 19) · The Hacker News (Oct 22) · TechCrunch, OpenAI raise (Oct 22) · TechCrunch, Meta ruling (Oct 21) · TechCrunch, Legora (Oct 21) · TechCrunch, Flapping Airplanes (Oct 21) · TechCrunch, Automattic (Oct 21)
