
A lone attacker armed with AI agents cracked over 100 online stores for roughly $50 a pop. Doctors may soon be able to gauge a cancer patient’s odds of survival from nothing more than a snapshot of their face. Businesses are waking up to five-figure AI invoices they never ran up, crooks are slipping invisible commands to AI assistants through everyday emails, and hundreds of supermarket freezers were remotely cranked up to 122 degrees. Every Tuesday this October, ThrottleNet owner George Rosenthal broke down the latest in tech on the Mark Reardon Show, 97.1 FM Talk. We’ve rounded up the five stories that grabbed our attention most, plus George’s practical advice for keeping your household and your company safe. Interested in learning more? Continue reading below.
A Single Attacker, a Fleet of AI Bots, and Over 100 Hacked Online Stores
Pulling off a large-scale cyberattack used to require a coordinated crew. These days, George explained, one individual can launch it with a click. Reports describe a lone hacker who turned a swarm of AI agents loose on online retailers across the U.S., Canada, Australia, and Japan. The bots went after more than 1,200 sites and got into over 100, about one in ten, with the attacker barely lifting a finger. The researcher who followed the attack pegged the price tag at around $50 for every site breached.
Retail sites make tempting targets since they hold customer data and handle card payments. George also noted that in September, researchers uncovered a different AI-driven operation that hit 25 businesses and walked off with no fewer than 600,000 credit card records.
What shoppers can do:
- Pay with credit, not debit, when buying online. A stolen credit card number is the bank’s problem to sort out. A stolen debit card number puts the cash in your checking account at risk.
- Set up instant transaction notifications from your bank so any odd charge gets flagged immediately.
What business owners can do: ask how long it’s been since your site was patched. Plenty of owners assume that refreshing pages and photos counts as an update, but the security that matters lives in the underlying platform and code that power the site. Confirm your IT provider is keeping that current.
George’s larger point is that criminals now have AI doing the heavy lifting, which means companies need AI on their side of the fight as well. Old-school security software can’t keep up on its own.
Can a Photo of Your Face Predict How Long You’ll Live?
This was the story that left everyone in the studio speechless. A team at Mass General Brigham had already built FaceAge, a tool that gauges a person’s biological age from a picture. Now they’ve gone further with a foundation model trained on over 40 million facial images. Its newest version, FAHR-FaceSurvival, estimates how likely a cancer patient is to die using just one photograph.
The AI zeroes in on spots like the temples, forehead, nose bridge, and the skin beneath the eyes, looking for frailty markers such as muscle wasting and loss of soft tissue. Researchers taught it using pictures of 30,000-plus cancer patients whose outcomes had already been recorded. In the group the model ranked as the top 10% for risk, roughly half died within three months.
George was impressed, but he also had real reservations:
- Insurers. Could insurance companies get their hands on these images and use them as grounds to reject claims?
- A growing archive of faces. Picture a future where your doctor snaps your photo at every appointment, just like stepping on the scale.
- Wrong answers. A mistaken prediction could put a family through needless grief, or lead doctors to order treatments a patient never needed.
- Oversight. George urged lawmakers to take up the issue right away.
In his view, there’s a big difference between AI reading an MRI or CT scan and AI looking at your face to forecast your future. The second one, he said, crosses into risky territory.
Stolen AI Logins Are Leaving Businesses With Massive Bills
Hackers have changed their playbook. In the past, breaking into a network usually meant encrypting files and holding them for ransom. Today, attackers first go looking for the passwords, usernames, and API keys connected to whatever AI tools the business relies on, whether that’s ChatGPT, Gemini, or Claude.
It all comes down to cost. AI platforms charge by the “token,” a measure of the computing power each prompt or task consumes. Once criminals have your login, they can run your account nonstop, using it to target other companies or to churn out malware. George shared that some businesses budgeting around $400 have been hit with $25,000 to $40,000 in usage charges over one weekend.
How to protect your company:
- Put firm spending limits on every AI service your team uses.
- Enable usage and billing notifications from your provider or your bank.
- Write an AI policy. When employees are using a patchwork of AI tools with no guidelines, leadership has no visibility. George puts an AI policy on the same level as ransomware defense, and ThrottleNet can help you put one together.
Invisible Commands: The New Way Scammers Trick Your AI Assistant
AI tools are shifting from talking to taking action. Ask a chatbot something and it gives you an answer. Hand an AI agent a job and it goes off and completes it by itself, such as scanning your inbox each Monday, responding to important contacts, and sorting everything else into folders.
Here’s the catch: an agent absorbs everything it’s sent to read, from emails to websites to files. George likened it to the professor who planted tiny white text in an assignment to catch students copying the prompt into AI. Fraudsters are doing the same thing, tucking commands into messages in microscopic type or in a color that blends into the background. You’ll never notice them, but your AI will. A buried instruction like “forward the last three bank statements” could be carried out by an agent that hasn’t been reined in.
These messages go out in huge volumes, and it only takes one victim whose agent can reach something valuable. George recommends:
- Make the agent ask first before anything sensitive. Explicitly forbid it from charging your card, sending email, or posting to group chats without your sign-off.
- Be specific about the boundaries. Agents are relentless about reaching their goal and may even recruit other agents to finish the job.
- Establish workplace AI rules before staff link agents to inboxes, financial data, or other company information.
This risk is already real. The New York Times reported on AI agents that, during internal testing, did things on live websites they were never meant to do, including completing visa paperwork on a State Department site and filing a bogus tip with a police department’s unsolved homicide portal. In another Times test, an AI agent navigating a basic maze smashed through walls it had never been taught to break because finishing the task paid off more than the penalties cost. The takeaway: a goal-driven agent will find a path, so you have to be the one setting the limits.
Melted Freezers, Outdated Firmware, and Quick Wins for Your Connected Home
In what George described as a jaw-dropping attack, cybercriminals took over the cooling controls at hundreds of supermarkets, most of them in France, and switched the refrigerators and freezers into defrost mode. Temperatures climbed as high as 122°F, and multiple grocery chains were forced to toss out ruined food.
What made it possible was old firmware, the built-in software that tells a device how to operate. The equipment maker said a current version would have blocked the break-in. Grocery chains aren’t the only ones exposed, either. Most homes now have plenty of internet-connected gear an attacker could exploit, from fridges and microwaves to thermostats, security cameras, doorbell cams, and smart TVs.
George suggests two easy steps:
- Switch on automatic firmware updates for all your connected devices. For anything that can’t update itself, check for updates periodically and install them manually. A tough Wi-Fi password won’t save you if the gadgets on your network are out of date.
- Move smart devices to a guest Wi-Fi network. Your cameras, TVs, thermostats, and appliances shouldn’t share a network with the laptop you work on. That way, a hacker who compromises one gadget can’t hop straight over to your computer. Sign in to your router, whether Spectrum, AT&T, or one you bought yourself, set up a separate guest network with its own password, and reconnect your devices to it.
Neither host knew guest networks were even an option, and that’s exactly why George says to take care of it today.
The Takeaway
AI is lowering the cost and speeding up the pace of cybercrime, and it’s showing up everywhere from the doctor’s office to the break room to the living room. Fortunately, a handful of smart habits go a long way: use credit instead of debit online, keep your website and smart devices patched, cap AI spending and adopt an AI policy at work, make AI agents get your approval before acting, and give your gadgets their own Wi-Fi network.
Tune in to Tech Talk Tuesday with George Rosenthal every week on the Mark Reardon Show, 97.1 FM Talk. Got a tech question you’d like George to tackle on air? Email [email protected] or contact us here!

Author: George Rosenthal – President & Co-Owner
