Hackers Grab Every Customer Login at Crypto-Crime Tracker Chainalysis, Investors Push Anthropic's IPO Price Down, and Microsoft Catches AI Agents Hunting Banks
By Noah Smith, Owner & Consultant, KeyChange Technologies · November 9, 2026
Your Monday AI and IT news covers Friday through Sunday, November 6 to 8, 2026: a crypto-crime tracker that lost every customer login, a lower price tag on the biggest AI IPO ever, AI hacking agents turned on banks, and a few tool changes worth five minutes of your week. If you missed Friday, catch up on the last recap here.
📌 The AI and IT news at a glance
- Chainalysis, the firm that traces crypto crime, says hackers exposed logins and live access tokens for all of its customers.
- Anthropic is reportedly ready to accept a lower valuation to get its IPO done before Thanksgiving.
- Microsoft says a Cambodia-based crime group is pointing an AI hacking agent at banks.
- A Korean court pauses a $148 million privacy fine against Coupang, but a bigger breach fine stands.
- Google takes over Crusoe's stalled Wyoming data center and plans to put $17 billion into it.
- Google sets an end date for the Gemini Code Assist perks bundled with Gemini Enterprise.
- Meta's Muse agent can now build mini planners and trackers called "Spaces."
- Amazon lets some US shoppers text Alexa to search, compare, and buy.
🔝 Top story
Hackers expose every customer login at Chainalysis, the company that tracks crypto crime
Chainalysis sells the software that crypto exchanges, financial firms, and investigators use to follow stolen money on the blockchain. Its own house is now the crime scene. The company first noticed an attack on October 30 and initially said only basic profile details such as names, emails, and role labels were touched. In an update late on November 4, which spread widely through the security press on Friday, it said the picture was much worse: an "on-path" attack, using what it called offensive agentic techniques, exposed names, usernames, password hashes, and live access tokens for all of its users.
Chainalysis reset every customer password, took several products offline (including Reactor and its Data Solutions service), and is bringing its transaction-monitoring and address-screening tools back in phases behind IP allowlists. Customers must reset passwords, rotate API keys, and give Chainalysis a list of trusted IP addresses to get back in. The company says it has no evidence that investigation data, such as case notes or attribution work, was accessed or altered. It has not named the attacker, and it has not said how many accounts were actually used by the intruder.
In short: Chainalysis says an attacker exposed password hashes and live access tokens for every one of its customers.
What it means for your business: A vendor breach can hand out working keys to your account even if your own systems are spotless. Live tokens matter more than passwords here, because a token can log someone in without ever needing the password or the MFA prompt.
My take: The lesson is not "crypto is risky." It is that every vendor login you hold is a live key somewhere else. When any SaaS vendor tells you to rotate keys, do it that day, then check that nothing new was created with the old ones. And if a vendor offers IP allowlisting, turn it on now, before you need it.
Source: Cybernews
🤖 AI
Anthropic reportedly trims its IPO price to get listed before Thanksgiving
Anthropic's public offering, expected to be the largest ever, ran into the tech selloff last week. Reuters reported on Thursday that several big investors were balking at a $1.5 trillion valuation, with one reportedly floating something closer to $1 trillion, and that the investor roadshow might slip. By Saturday, the Financial Times reported that Anthropic is pushing ahead anyway and will accept a lower price: it is now aiming to raise $80 billion to $100 billion at a valuation between $1.25 trillion and $1.75 trillion, and to go public before Thanksgiving.
For context, Anthropic's last private valuation was $965 billion, and it is running at roughly $81 billion in annualized revenue. The pullback came as the Nasdaq 100 fell 6.4% on Thursday, part of a broader slide in AI stocks after the midterm elections raised the odds of tighter rules on data centers.
In short: Anthropic is reportedly willing to take a lower valuation, between $1.25 trillion and $1.75 trillion, to complete its IPO before Thanksgiving.
What it means for your business: Nothing changes in the Claude products you use today. A public Anthropic will face quarterly pressure on pricing and growth, which tends to show up later as plan changes, new tiers, and price moves.
My take: A trillion-dollar company haggling over a few hundred billion is a reminder that AI pricing is still being figured out on the fly. Keep your AI tools on month-to-month terms where you can, and avoid building critical workflows that only work with one vendor.
Source: The Decoder, citing the Financial Times
Google takes over Crusoe's troubled Wyoming data center and plans to spend $17 billion on it
Crusoe, the AI data center builder valued at $30.9 billion, confirmed on Friday that it sold an unfinished data center project in southeastern Wyoming to Google. The price was not disclosed. Google, which reportedly had concerns about the project's cost and timeline, plans to invest $17 billion in the campus. The site was originally announced with energy company Tallgrass with an eventual goal of 10 gigawatts of capacity.
Crusoe is not walking away from Google entirely: it is still building a separate campus for Google near Amarillo, Texas. The deal lands in the same week that the AI stock selloff and the midterm results put new scrutiny on how fast these giant projects can actually get built.
In short: Google bought Crusoe's unfinished Wyoming data center project and plans to put $17 billion into it.
What it means for your business: The AI services you rent run on a handful of physical sites, and the big cloud providers are taking more of that capacity under their own roof. That tends to mean more stable supply from the giants and a harder road for smaller AI hosts.
My take: When even a $30 billion builder has to hand a project to Google, it tells you how hard the physical side of AI really is. For a small business, the practical read is to favor AI tools that run on the major clouds, because that is where the capacity is landing.
Source: TechCrunch
🛡️ IT and security
Microsoft says a crime group is aiming an AI hacking agent at banks
Microsoft's threat intelligence team said on Friday that a financially motivated group it tracks as Storm-3358, based in Cambodia, has been using ARTEX, an agentic AI hacking framework, against financial organizations since October. Most targets are in Southeast Asia, but Microsoft says a smaller number of large global financial firms are on the list too. ARTEX automates the boring parts of an attack: scanning for weak spots, exploiting them, dropping remote-access tools, stealing credentials, and moving to other machines.
Microsoft's advice is telling. It recommends decoys, honeypots, and "tarpits" designed to waste an AI agent's time, plus risk-based sign-in rules in Microsoft Entra ID. It also warns that simply blocking AI agents by their browser signature will not work, because these tools can pretend to be anything.
In short: Microsoft says a Cambodia-based crime group has been using an AI hacking agent called ARTEX against banks and other financial firms since October.
What it means for your business: AI agents make it cheap for criminals to probe thousands of companies at once, so being "too small to target" is less true every month. The basics (MFA everywhere, patched internet-facing systems, risk-based sign-in rules) are what slow these tools down.
My take: This is the one technical story I'd ask you to read this week, because it is the shift in one paragraph: attackers no longer need a skilled human at the keyboard for the early stages. If you use Microsoft 365, ask whoever manages it whether conditional access is turned on. That one setting does a lot of work against automated logins.
Source: Microsoft Security Blog
A Korean court pauses a $148 million privacy fine against Coupang, but a bigger breach fine stands
Coupang, South Korea's largest online retailer, won a temporary reprieve on Friday. The Seoul Administrative Court suspended a 201.1 billion won (about $147.7 million) fine from Korea's privacy regulator, which said Coupang collected and stored the online activity records of about 11.17 million users without their consent. The fine is on hold until 30 days after the first verdict in Coupang's lawsuit challenging it, with the court citing possible irrecoverable damage to the company.
The relief only goes so far. A day earlier, a different panel of the same court rejected Coupang's request to pause a separate 423.6 billion won fine over a data breach affecting more than 37 million users, which is the regulator's largest fine ever against a single company.
In short: A Seoul court suspended a $148 million privacy fine against Coupang while it fights the penalty, but refused to pause a larger fine over a 37-million-user breach.
What it means for your business: Regulators are now pricing privacy failures in the hundreds of millions, and "we didn't get consent" can cost as much as "we got hacked." Know what customer data you collect, why, and whether your consent language actually covers it.
My take: The headline is the pause, but the real story is the breach fine that stuck. Courts may debate how you used data; they are far less sympathetic when you lost it. Encrypt what you store and delete what you no longer need.
Source: The Korea Times
🧰 Tools for vibe coders and business AI
Google sets an end date for the Code Assist perk bundled with Gemini Enterprise
In Friday's Gemini Enterprise release notes, Google said the Gemini Code Assist entitlements bundled with existing Gemini Enterprise Standard and Plus subscriptions will end on March 31, 2027. Until then, that bundled access is limited to the Gemini 3.x model family. New subscriptions have not included Code Assist since October 9, 2026.
Google's answer is Antigravity, its agentic coding tool, which it says is included in Gemini Enterprise editions. In other words, the coding help is not disappearing, but it is moving to a different product with a different workflow.
In short: Google will end bundled Gemini Code Assist access for existing Gemini Enterprise Standard and Plus customers on March 31, 2027, and points them to Antigravity instead.
What it means for your business: If your team codes with Gemini Code Assist through a Gemini Enterprise plan, you have about four and a half months to test Antigravity and move over.
My take: Put the date on a calendar now and run a two-week trial of Antigravity in the new year. Migrations are painless when they are planned and miserable when they are discovered the week of the cutoff.
Source: Google Cloud: Gemini Enterprise release notes
Meta's Muse agent can now build mini planners and trackers called "Spaces"
Meta added "Spaces" to Muse, its AI agent, on Friday. Spaces are breakout areas in the Muse sidebar that work like small custom apps: you describe what you want in chat, such as a vacation plan, a holiday gift tracker, a wedding planner, or a DIY project, and Muse builds it. Connected third-party apps keep each space updated in the background.
Mark Zuckerberg announced the feature on X. It follows Thursday's launch of Muse's real-time voice mode, and Meta said in October that Muse agents had about 10 million users in the US and Canada.
In short: Meta launched Spaces, which lets Muse turn a chat request into a small, self-updating planner or tracker.
What it means for your business: "Describe it and the AI builds a little app" is now a consumer feature, not just a developer tool. Expect customers and staff to show up with homemade trackers built this way.
My take: Handy for personal projects. For anything involving customer or company data, keep it in tools your business actually controls, because a Space connected to someone's personal accounts is not where client information belongs.
Source: TechCrunch
Amazon lets some US shoppers text Alexa to search, compare, and buy
Amazon announced on Friday that select US customers can now shop by texting Alexa. Invited customers get a text to turn it on, and everyone else can join a waitlist, with a broader rollout planned in the coming weeks. Over text, Alexa can search for products, add them to a cart, make a purchase, manage lists, track prices, compare products, check order status, and search by image.
The move comes two months after Amazon blocked Meta's Muse agent from querying its site, which makes the strategy clear: Amazon wants shoppers using its own AI assistant, not someone else's.
In short: Amazon is rolling out texting with Alexa for shopping, starting with select US customers.
What it means for your business: If you sell on Amazon, more of your buyers will find you through an AI assistant rather than a search results page. Clear titles, specs, and photos are what that assistant reads.
My take: The big retailers are building walls around their AI shoppers. If you sell online, make sure your product data is clean and complete everywhere you list, because the assistant only recommends what it can understand.
Source: TechCrunch
That's the weekend's AI and IT news. The thread running through it: your vendors' security is your security, and the AI tools you rely on are still changing price, packaging, and plumbing. Rotate the keys you were asked to rotate, check your Microsoft 365 sign-in rules, and calendar the March 31 Gemini deadline if it applies to you.