A Cyberattack Diverts Ambulances From New Mexico's Only Top Trauma Center, AI Agents Slip Into Anthropic Through Its Vendors, and OpenAI Halves the Price of Its Desktop-Driving Model

By Noah Smith, Owner & Consultant, KeyChange Technologies · October 27, 2026

Here is your AI and IT news for Tuesday, October 27, 2026. A hospital system went offline, a startup's AI agents found a back door into Anthropic through ordinary business software, and OpenAI cut the price of the model built to run your desktop. If you missed yesterday's roundup, catch up on the previous AI and IT news recap.

📌 The AI and IT news at a glance

• 🔝 A cyberattack forces New Mexico's largest hospital to divert ambulances and cancel every clinic appointment.
• 🤖 OpenAI launches GPT-6.1 Terra, a cheaper, faster model built to click through forms and desktop apps for you.
• 🤖 Meta plans to sell up to $95 billion in stock to pay for AI data centers.
• 🤖 Microsoft offers voluntary buyouts to about 60,000 employees as it reshapes for AI.
• 🛡️ A startup's AI agents got into Anthropic's internal systems in under 12 hours, through third-party business software.
• 🛡️ Korean police say a former pentester's gang used an AI hacking suite to steal 8.2 million records from 90+ organizations.

🔝 Top story

Cyberattack shuts down UNM Health: ambulances diverted, clinics closed statewide

On Monday, the University of New Mexico Health System diverted ambulances to other Albuquerque hospitals and canceled all appointments and outpatient surgeries at its more than 30 clinics across the state after a cybersecurity threat hit its computer systems. UNM Hospital is New Mexico's largest hospital and its only Level 1 trauma center. Technicians rebooted the network in the early afternoon, and the system said it began restoring operations in phases "as it was safe to do so." Students and nonessential staff across the Health Sciences Center were sent home, and Health Sciences classes and research were canceled again for Tuesday.

UNM has not said whether this is ransomware or whether patient data was taken, and it calls the matter an active law enforcement investigation. Federal and state law enforcement were notified, and Sen. Martin Heinrich said he was working to bring in federal resources. For context, the Albuquerque Journal cites a Comparitech estimate that healthcare organizations average about 17 days of downtime after a cyberattack.

In short: A cyberattack knocked UNM Health's systems offline Monday, forcing ambulance diversions and canceled appointments statewide.

What it means for your business: When a big organization goes down, everyone who depends on it feels it, including its vendors, patients, and neighbors. Know which outside services your business cannot operate without, and what you would do on the day one of them goes dark.

My take: The detail that sticks with me is "rebooted the network." Even with the best people, the first move in a real incident is often to pull the plug and restore in pieces. That only works if you have practiced it. If your plan for a bad day has never been tested, it is a hope, not a plan.

Source: Albuquerque Journal

🤖 AI

OpenAI launches GPT-6.1 Terra: half the price, built to run your desktop

OpenAI released GPT-6.1 Terra on Monday, a fast, lower-cost model aimed at computer use: filling out forms, working through desktop apps, reading technical drawings, and automating repetitive screen work. API pricing is $1 per million input tokens and $5 per million output tokens, which OpenAI says is 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol. OpenAI also says Terra is more than three times faster than GPT-6 Astra and GPT-6.1 Sol.

The headline numbers are about reliability. On OpenAI's internal computer-use benchmark, the error rate at low reasoning effort dropped from 53.6% on GPT-5.6 Terra to 16.7%, and at maximum effort it fell to 4.3%. Terra is available now to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and in the API. OpenAI still points people to Sol or Astra for the hardest coding work.

In short: OpenAI shipped a cheaper, faster model designed specifically to operate software on a computer the way a person would.

What it means for your business: The boring, clicky tasks (data entry across two systems, filling the same portal form every week) are getting cheaper and more reliable to hand to AI. It is worth listing those tasks now so you know where to test first.

My take: An error rate dropping from roughly one in two to roughly one in six is real progress, but one in six is still a lot if the task touches money or customers. Start with low-stakes, easy-to-check work, and keep a human reviewing the output until it earns your trust.

Source: BleepingComputer

Meta plans to sell up to $95 billion in stock to fund AI data centers

Meta announced Monday that it will sell up to $95 billion in stock to pay for its AI infrastructure buildout. The first phase is $35 billion: roughly $23 billion in common stock (Nvidia, AMD, and Broadcom have expressed interest in buying up to $5 billion of it) plus $12.5 billion in mandatory convertible preferred shares. A follow-on "at-the-market" sale of up to $60 billion is expected to start early next year.

The money goes to data centers, including four AI "superclusters" of more than a gigawatt each. Meta's Hyperion campus in Louisiana is slated for 5 gigawatts and will need 10 new power plants. It is a striking shift for a company that has spent years buying back its own shares rather than selling new ones.

In short: Meta is raising up to $95 billion from the stock market to keep building AI data centers.

What it means for your business: The AI tools you use run on infrastructure this expensive, and someone eventually pays for it. Expect pricing for AI features to keep shifting, and avoid locking into long contracts on the assumption that today's prices are permanent.

My take: When the richest companies on earth have to raise outside money to keep building, it tells you how big the bet is. That does not mean AI is a bubble, but it does mean the vendors need you to pay up eventually. Use the cheap prices while they last, and keep your options open.

Source: SiliconANGLE

Microsoft offers buyouts to about 60,000 employees

Microsoft confirmed Monday that it is offering voluntary exit packages to roughly 60,000 employees, at vice president level and below, on select teams in the U.S., Canada, India, and the U.K. Bloomberg reports that Xbox, Windows, Copilot, and Finance are among the participating divisions, while data center and sales and marketing staff are excluded. Microsoft had about 223,000 employees at the end of June. The terms are described as similar to its standard severance.

Chief People Officer Amy Coleman framed it as a way for people already considering a move to leave with support. TechCrunch describes it as part of Microsoft's effort to cut costs and rework its workforce for the AI era. An earlier, smaller buyout this year was offered to about 8,750 employees, and more than 30% accepted.

In short: Microsoft is offering voluntary buyouts to about a quarter of its workforce as it shifts spending toward AI.

What it means for your business: The vendors behind Windows, Copilot, and Microsoft 365 are reorganizing. In the short run, that can mean slower support and shifting product priorities, so do not count on a feature you need arriving on schedule.

My take: Buyouts are the polite version of layoffs, and they tell you where a company's money is headed. Microsoft is spending heavily on AI infrastructure and trimming everywhere else. If you rely on a Microsoft product that is not at the center of that strategy, keep an eye on it.

Source: TechCrunch

🛡️ IT and security

AI agents got into Anthropic in under 12 hours, through its vendors' software

Cathedral, a months-old security startup founded by former DOGE staffers, says its AI agents broke into Anthropic's internal systems in under 12 hours, according to a draft report reviewed by the Washington Post. The way in was not Anthropic's own code. Cathedral's agents spent days failing to breach Anthropic directly, then succeeded through flaws in third-party administrative and sales software that many other companies also use. Cathedral's draft report says the agents found paths to meeting transcripts, salary and job-offer details, and Slack messages.

Anthropic's account is narrower. It says the researchers reached a "limited set" of sales, HR, and IT systems, including business contact information and basic user account records, and did not access Claude users' conversations, source code, or model weights. Anthropic says its monitoring flagged some of the activity, it cut off access the same day, it disabled the affected software until the vendors confirmed fixes, and it has no evidence anyone else exploited the flaws. The testing happened under a bug-bounty-style program, and Anthropic accuses Cathedral of breaking the rules by not disclosing the full extent of its activity. The vendors have not been named.

In short: A startup's AI agents breached a limited set of Anthropic's internal business systems by going through third-party software rather than Anthropic's own defenses.

What it means for your business: Your security is only as strong as the HR, sales, and admin tools you plug into. Know which outside apps hold your staff and customer data, and turn off the ones you no longer use.

My take: The lesson here is not about Anthropic. A well-funded AI company with a serious security team got in through the side door: everyday vendor software. Most small businesses have far more of those side doors and far less monitoring. Take inventory of what is connected to your accounts this week.

Source: Decrypt

Police say a former pentester's gang used an AI hacking suite to steal 8.2 million records

South Korean police have identified seven suspects in a cybercrime ring that breached more than 90 organizations, including schools, hospitals, large corporations, and golf courses, and stole an estimated 8.2 million records. The alleged leader is a 32-year-old who worked as a penetration tester at a cybersecurity company from 2019 to 2022. He was arrested October 15 and referred to prosecutors October 23. The other six were arrested but not held in custody.

Police say the group started with conventional hacking tools in September 2025 and switched to AI tools in late August 2026, mostly an autonomous attack framework called ARTEX AI, which has also been tied to attacks on South Korean banks. In 38 cases, the gang demanded a combined 360 million won (about $270,000) not to leak stolen data. Investigators are still checking whether any of it surfaced on the dark web.

In short: A small gang allegedly used an AI hacking tool to breach 90+ organizations and steal 8.2 million records.

What it means for your business: AI tools let a handful of people attack dozens of organizations at once, so being "too small to target" is no longer a defense. The basics (patching, strong login protection, offline backups) matter more, not less.

My take: Seven people, 90-plus victims. That ratio is the story. AI is not inventing new kinds of attacks so much as letting small crews work like large ones. The good news is that the same basic defenses still stop most of them.

Source: BleepingComputer

That is the AI and IT news for October 27. See you tomorrow.