Meta Is Ordered to Pay $25 Billion Over Broken Privacy Promises, Capcom Exposes 11.7 Million Accounts, and Anthropic Unveils an AI Assistant That Messages You First
By Noah Smith, Owner & Consultant, KeyChange Technologies · October 23, 2026
Here is your AI and IT news for Friday, October 23, 2026: a $25 billion privacy penalty for Meta, a breach at a gaming giant that touches 11.7 million inboxes, an AI-driven attack on Microsoft 365 tenants finally gets a name, and Anthropic launches an assistant that messages you first. Everything below broke in the past two days, verified, and written for owners who want the signal without the jargon. (Catch up on the previous edition here: Wednesday's AI and IT news recap.)
📌 The AI and IT news at a glance
- 🔝 A judge orders Meta to pay New Mexico $25 billion for misleading users about their privacy.
- 🤖 Anthropic unveils Claude Duet, an AI assistant that checks your apps and reaches out first.
- 🤖 OpenAI is reportedly raising $30 billion at a record $1.25 trillion valuation.
- 🤖 Walmart opens its store to AI shopping agents, with spending caps, while Amazon blocks them.
- 🛡️ Capcom confirms a breach exposing 11.7 million customer email addresses.
- 🛡️ Microsoft names the China-aligned group behind the ConnectWise backup hack, and it ran on AI agents.
- 🛡️ A cyberattack knocks phones, email, and Wi-Fi offline at all 53 Richmond public schools.
- 🛡️ Interpol arrests 1,462 people and recovers $105 million in a sweep of African cybercrime rings.
- 🧰 Google Meet's live speech translation now covers 63 languages and detects them automatically.
- 🧰 Railway, a popular hosting platform for app builders, tells customers to rotate their secrets after a breach.
🔝 Top story: Meta ordered to pay $25 billion over privacy promises it did not keep
On Wednesday, New Mexico District Judge Francis Mathew ordered Meta to pay the state $25 billion for violating its Unfair Practices Act. The penalty follows a September jury verdict in Santa Fe that found Facebook committed close to 44 million violations, siding with the state on 31 of 34 counts. The claims centered on misrepresentations about how much control users had over their data, Meta's handling of hate speech and misinformation, and the investigation that followed the Cambridge Analytica scandal. The law allows up to $5,000 per violation, so the state was technically entitled to nearly $220 billion; the attorney general asked for up to $40 billion. The judge pointed to "the clearly intentional and systematic nature" of the misstatements and the involvement of senior leadership as reasons for the stiff penalty. Unpaid amounts accrue 15% annual interest.
Meta's chief legal officer called the ruling something that makes it "all but impossible to do business" in New Mexico, and the company says it is confident it will win on appeal. Attorney General Raúl Torrez noted the figure tops the multistate settlements reached earlier this year.
In short: A New Mexico judge ordered Meta to pay $25 billion after a jury found tens of millions of deceptive statements about user privacy and data control.
What it means for your business: This was not a hacking case; it was a "you said one thing and did another" case, and consumer-protection laws let states multiply small per-violation fines by every affected customer. What your privacy policy, website, and sales team promise about customer data is a legal commitment, not marketing copy.
My take: Most small businesses will never see a nine-figure penalty, but the math behind this one is the lesson: a modest fine times every customer adds up fast. Pull up your privacy policy this week and ask one question: does it describe what we actually do with data today, including the AI tools we have added since it was written? If not, fix the policy or fix the practice.
Source: Source New Mexico: New Mexico judge orders Meta to pay state $25B
🤖 AI generally
Anthropic launches Claude Duet, an assistant that reaches out before you ask
Anthropic introduced Claude Duet, a proactive AI agent built on Opus 5.5, a model it has not yet released on its own. Instead of waiting for a prompt, Duet connects to your apps, sends a short morning brief on what needs attention, then checks in every few hours and pings you when something comes up. It can handle simple admin chores or larger jobs like building a small web app, and it runs in the cloud on its own virtual machine with its own browser, so it does not need access to your computer. It reuses the memory, skills, and app connections already in your Claude account, and everything happens in one ongoing thread.
Duet is a research preview available by waitlist to Pro and Max subscribers in the US, on the web and in a standalone iOS app. There is no separate price; its background checks count against your existing plan limits. Anthropic says Duet asks for approval before sending messages, making payments, signing documents, or deleting files, and that it never sees your passwords when given account access. It lands in a crowded field: Meta's Muse and OpenAI's Dots are pitched at the same "always-on assistant" idea.
In short: Anthropic opened a waitlist for Claude Duet, an Opus 5.5-powered agent that monitors your connected apps and proactively flags or handles work.
What it means for your business: "Assistants that work between messages" are now a real product category from all three major AI labs, aimed squarely at small business owners juggling too many inboxes. The question is shifting from "can AI help" to "which of my accounts am I comfortable letting it watch."
My take: The approval-before-action design is the right default, and I would keep it on. If you try Duet, start by connecting one low-stakes app, like a calendar or a project board, and watch what it surfaces for a week before handing it email or anything that touches money. Also keep an eye on your plan limits, since background checks quietly use them up.
Source: MacRumors: Anthropic Launches Claude Duet, a Proactive AI Agent
OpenAI is reportedly raising $30 billion at a $1.25 trillion valuation
According to unnamed sources cited by CNBC and summarized by TechCrunch, OpenAI is negotiating a new raise of up to $30 billion at a $1.25 trillion pre-money valuation, which would be a record for a private company. Coatue and investors from the UAE are said to be leading, possibly alongside a deal for more computing capacity from the UAE. OpenAI declined to comment, and the terms are still being negotiated. For context, OpenAI's last raise in March valued it at $852 billion, and it filed confidentially for an IPO in June. Anthropic, its closest rival, reportedly wants a $1.5 to $2 trillion valuation for its own IPO.
In short: OpenAI is reportedly in talks to raise up to $30 billion at a $1.25 trillion valuation ahead of a public listing; the company has not confirmed it.
What it means for your business: The money pouring into the top AI labs is what funds the compute behind the tools you use, but these companies will eventually face public-market pressure to turn that spending into profit. Expect pricing, plan tiers, and ads (OpenAI just expanded ChatGPT ads to 89 countries on its free tiers) to keep changing.
My take: Treat this as weather, not news you need to act on. The practical takeaway is to avoid building anything critical around one vendor's current pricing. If a tool is central to your operations, know what it would take to switch.
Source: TechCrunch: OpenAI working on new $30B raise at $1.25T valuation ahead of IPO
Walmart welcomes AI shopping agents while Amazon shuts them out
Walmart CEO John Furner announced Agent Gateway, a new server that lets approved AI agents (Walmart already works with Meta's Muse and offers integrations with ChatGPT and Gemini) search its inventory and buy on a customer's behalf, at both Walmart and Sam's Club. Starting next month, Agent Profiles will give each agent its own space inside the customer's account, so its purchases show up separately, much like adding an authorized user to a credit card. Customers can require approval for every purchase or set a spending limit, and they can delete an agent's profile at any time.
Walmart's rules say agents cannot see payment information, can only access what a task needs, and will be blocked if they break the terms; normal returns apply if an agent gets an order wrong. No Walmart+ membership is required. The move is a sharp contrast with Amazon, which has started blocking AI agents from purchasing on its site, including Meta's.
In short: Walmart launched a sanctioned doorway for AI agents to shop for customers, with per-agent profiles, spending caps, and approval controls.
What it means for your business: "Agentic commerce" just got its first big-retail playbook: let agents in, but give them their own identity, a budget, and a leash. If you sell online, some of your future customers will be software acting for a person, and you will need to decide whether to block them or welcome them on your terms.
My take: The spending-limit and separate-profile model is exactly how I would want any business to let AI touch purchasing, including your own internal purchasing. If you are letting an assistant order supplies, give it its own account and a hard cap rather than your company card.
Source: TechCrunch: While Amazon blocks AI agents, Walmart rolls out the welcome mat
🛡️ IT and security
Capcom confirms a breach exposing 11.7 million customer email addresses
Capcom, the company behind Street Fighter and Resident Evil, confirmed that an outsider broke into its CAPCOM ID membership service and exposed "more than 11.7 million customer email addresses." Along with emails, the attackers got account creation dates, last login times, login counts, and the IP address used at last login. For 43 accounts, birthdates, gender, nicknames, and countries were also taken. For 196 accounts, the attackers went further and changed passwords to lock the owners out; Capcom has reset those. The company says other passwords and all credit card data were not exposed.
Capcom spotted the intrusion on October 19, confirmed customer data was involved the next day, and reported it to police and Japan's privacy regulator on October 21. It says it has identified how the attackers got in and added protections, but has not shared details, and it will notify affected customers in stages.
In short: Capcom says a breach of its CAPCOM ID service exposed more than 11.7 million email addresses plus login details, and 196 accounts were hijacked.
What it means for your business: A list of 11.7 million real, active email addresses with recent login times is prime material for phishing, and gamers include plenty of your employees. Expect convincing "Capcom account" emails, and the same addresses reused against other services.
My take: "Only email addresses" is never as harmless as it sounds; it is the starting list for the next phishing wave. This is a good week to remind staff that a breach notice email is itself a favorite disguise, and to go straight to the company's site rather than clicking a link.
Source: BleepingComputer: Capcom confirms data breach affecting 11.7 million accounts
Microsoft names the AI-assisted group behind the ConnectWise backup hack
Earlier this week we covered how a compromised ConnectWise cloud backup service became a doorway into customers' Microsoft 365 accounts. Microsoft has now attributed that attack to Storm-0940, a China state-aligned group, and the details show how much AI sped things up. The group used ARTEX, an agentic AI hacking framework, to automate scanning, password spraying, and abuse of connected apps. Between October 10 and 14, it fired off password-guessing attempts against more than 2,250 Microsoft 365 tenants, peaking at 2,900 sign-in attempts per minute. On October 15, the AI agents took control of identities belonging to several ConnectWise cloud apps and used that trust to sign into customer environments.
In several cases, less than three hours passed between first targeting and privileged access to cloud infrastructure. Once inside, the attackers changed app credentials so they could get back in later, and altered users, groups, and admin roles. Zscaler found prompts in the attackers' tooling asking for bulk personal and transaction data and payment-gateway keys, which points to a financial motive. ConnectWise shut down services and is restoring them in phases.
In short: Microsoft tied the ConnectWise supply-chain attack to China-aligned Storm-0940, which used AI agents to go from first contact to admin access in under three hours.
What it means for your business: If an IT provider or app connects to your Microsoft 365, its access is your access. And with AI doing the legwork, the window to catch an intrusion is now measured in hours, not days.
My take: This is the one technical story I would make sure your IT person reads. Ask them two things: which third-party apps have permissions in our Microsoft 365, and would we notice if one of them suddenly got new credentials? An attacker who plants a new app secret can stay in even after the original hole is closed.
Source: The Hacker News: China-Linked Hackers Use ARTEX AI Agents in ConnectWise Supply Chain Attack
A cyberattack takes phones, email, and Wi-Fi offline at all 53 Richmond public schools
Richmond Public Schools in Virginia shut off internet access districtwide on Wednesday after employees saw a warning on their district laptops claiming sensitive data had been stolen and threatening to publish it unless the district made contact. The incident disrupted phones, email, and online learning tools at all 53 campuses serving about 22,000 students, and sources told local media it also hit Wi-Fi, cameras, and the PowerSchool portal parents use for grades and attendance. Classes continued, with teachers adjusting on the fly.
Important caveat: the district has not confirmed the message's claims. It has not said whether data was actually stolen, whether files were encrypted, or whether this is ransomware, and no ransomware group has publicly taken credit. RPS is working with outside experts and law enforcement and has not said how long the shutdown will last.
In short: An unconfirmed extortion message on staff laptops led Richmond Public Schools to pull its network offline across 53 schools.
What it means for your business: The first hours of an incident look like this: a scary message, no confirmed facts, and a decision to disconnect. Having that decision made in advance, and a way to keep working with phones and email down, is what separates a bad day from a lost week.
My take: Pulling the plug quickly was a defensible call, and the district is being careful not to overstate what it knows. Ask yourself: if your network had to go dark at 9 a.m. tomorrow, how would your team call customers, and who decides to pull the plug?
Source: Hoodline: Cyberattack Knocks Out Phones, Email at 53 Richmond Public Schools
Interpol arrests 1,462 suspects and recovers $105 million in an African cybercrime sweep
Interpol announced the results of Operation Serengeti III, which ran from June 29 to August 31 across 21 countries, including the United States. Police arrested 1,462 suspects, disrupted 10,909 criminal networks, and recovered more than $105 million. Investigators identified 65,929 victims and linked the activity to losses of more than $385 million. The haul included investment and pension fraud rings in Uganda that led to 365 arrests, illegal crypto mining centers in Angola, and a phishing and business email compromise operation in Nigeria that had collected about 2.1 million email addresses.
In short: A two-month, 21-country Interpol operation arrested 1,462 cybercrime suspects and recovered over $105 million.
What it means for your business: Many of these crews run exactly the scams that hit small businesses: fake invoices, hijacked email threads, and too-good-to-be-true investment pitches. Arrests help, but the playbooks live on with whoever is still free.
My take: Nice to see real numbers on the enforcement side. The defense that would have stopped most of these victims is still the cheap one: a phone call to a known number before you change payment details or wire money.
🧰 Tools for vibe coders and everyday business AI
Google Meet's live translation now covers 63 languages and figures out who is speaking what
Google expanded Gemini-powered real-time speech translation in Meet to 63 languages, and it now detects the spoken language automatically and translates it for each listener. Each participant turns it on for themselves. It is available on Workspace Business Standard and Plus, Enterprise Standard and Plus, Google AI Pro and Ultra, and a few other plans. Base Workspace and AI Pro plans get up to 90 minutes per meeting, while higher tiers get up to 8 hours. It does not work in live streams, recordings, or client-side encrypted meetings.
In short: Google Meet's live voice translation expanded to 63 languages with automatic language detection on most paid Workspace plans.
What it means for your business: If you work with overseas suppliers, contractors, or customers, a decent interpreter may already be included in the plan you pay for.
My take: Worth a test run on a low-stakes call. I would not rely on it for contract terms or anything legal, but for routine check-ins it can save a lot of awkward back-and-forth.
Source: Google Workspace Updates: Real-time speech translation in Google Meet expands to 63 languages
Railway tells customers to rotate their secrets after a breach
Railway, a hosting platform popular with people building and launching apps quickly, disclosed "anomalous activity" in the systems that store customers' environment variables, which is where apps keep their passwords and API keys. An intruder may have accessed those encrypted values and, for some customers, their decrypted contents. A security researcher reported that attackers drained cryptocurrency wallets whose keys were stored on Railway, with the first known theft at least two hours before the public disclosure. Railway says it rotated its own credentials, invalidated API tokens, and alerted upstream providers, and believes the attack is part of a broader campaign aimed at platform companies on the way to AI providers.
In short: Railway says an intruder may have read customer app secrets and is urging everyone to rotate API keys and credentials.
What it means for your business: If someone built an app, internal tool, or website for you on Railway, the keys to your other services (payments, email, AI APIs) may have been exposed.
My take: If you have anything on Railway, rotate every key stored there today; do not wait for a notice. And this is a good reminder for every vibe coder: keep secrets in one place you can rotate quickly, and never keep crypto wallet keys in an app's settings.
Source: The Hacker News: Railway Says Intruder May Have Accessed Customer Secrets, Urges Credential Rotation
That is the AI and IT news for this edition. If one item deserves an action today, make it the ConnectWise check: know which outside apps can sign into your Microsoft 365, and look again.