You What? The AI, Privacy, and Security Weekly Update.
EP 314
In this week's update,
Imagine an AI agent running out of test cryptocurrency, finding a stranger's wallet, and helping itself.
Imagine a single photo of your bathroom being enough for police to figure out your address.
Imagine a hacker using AI agents to reach more than 100 companies for about $5,000, and a federal agency being asked to hack a voting machine and keep the results in its back pocket.
That is the reaction this week keeps producing. You what?
Some of it is funny. Silicon Valley's enthusiasm for AI is so intense that, according to Palantir's CTO, some people have started smoking, betting that a machine will cure cancer before the cigarettes kill them.
Some of it is alarming. A Florida driver's license database and a California student records system were both breached through login credentials that should never have been usable by an outsider.
And some of it is simply hard to believe: a state tried to require websites to do the impossible, and a court had to step in.
Nine stories. Surveillance, AI agents behaving badly, breaches, bad bets, and a few institutions asking the public to simply trust them.
Let's start with the claim that stopped us in our tracks.
Global: Show You Believe, Let's Go for a Smoke
Apparently, some people in Silicon Valley believe in AI so much that they've started smoking cigarettes, confident the technology will cure cancer before it catches up with them. That's the claim made by Shyam Sankar, CTO of Palantir, during a CBS News interview. He offered no names or evidence to back it up, but the idea fits a growing strain of thinking in tech circles: bet everything on AI because it will eventually solve whatever problems we create along the way.
Smoking has also become a social accessory for a small group of tech founders and investors. Some see cigarette breaks as networking opportunities or a way to appear less polished. Others genuinely believe AI will eliminate the health risks before those risks catch up with them. One startup founder reportedly suggested that by the time smokers develop lung cancer, AI-powered robots will have cured it.
Sankar's bigger concern is that Silicon Valley has developed something close to religious faith in artificial general intelligence. Some people see AI as humanity's salvation, while others fear it could destroy us. Either way, people are placing extraordinary faith in a technology whose ultimate capabilities and consequences remain uncertain.
There's a serious problem with both extremes. AI could make enormous advances in medicine, but that doesn't mean every disease will be cured or that breakthroughs will arrive on our personal schedules. And while AI can help researchers discover treatments, it hasn't made cigarettes safe.
So what's the upshot for you?
Enjoy the benefits of AI, but don't bet your health, finances, or future on predictions from people selling the technology. If Silicon Valley's brightest minds are willing to smoke today because a machine might cure them tomorrow, that's a pretty good reminder to keep your own risk management human.
If the people selling AI are betting their lungs on it, imagine what else they might be willing to gamble with. Start with the photos on your phone. A tool called Raven can now tell police where a picture was taken from nothing but the inside of a room.
US: AI Software Allows Police to Geolocate You Using a Single Interior Image
Forget GPS tracking. A new AI tool called Raven can potentially figure out exactly where you are from a single photograph taken inside a building. A picture of a bathroom, washing machine, or even a hotel toiletry bottle could be enough.
The system, developed by Graylark Technologies, compares small visual details against millions of online property listings to identify the location. Unlike a traditional reverse image search, Raven doesn't need to find an identical photograph. It looks for matching features such as tiles, toilets, appliances, and other details that could identify a specific property. It apparently doesn't need GPS coordinates or image metadata, either.
According to reporting by 404 Media, the Los Angeles Police Department and Miami-Dade Sheriff's Office have purchased access to Raven. The technology may also have helped investigators locate influencer Andrew Tate, who was arrested in Miami in July 2026, although authorities have not confirmed that Raven was used in his case.
The implications extend well beyond law enforcement. A photograph posted from a hotel room, vacation rental, or private residence could reveal far more than its owner intended. Even if you switch off location tagging, the room itself may provide enough clues to identify the property.
So what's the upshot for you?
Before posting photos or videos, check what's visible behind you, especially in hotel rooms and vacation rentals. Crop out distinctive fixtures, appliances, and other identifying details when possible. Turning off GPS tagging is useful, but it won't hide a recognizable bathroom. In the age of AI, you might be giving away your address without ever showing your front door.
Your bathroom tiles can now give away your address. But an AI tool doesn't need to get into your home to cause trouble. Sometimes it just needs access to the wrong wallet, which is exactly what happened when an OpenAI agent ran out of test crypto and went shopping.
Global: AI Needed Crypto, So It Took Someone Else's
An OpenAI AI agent ran out of cryptocurrency while working on a coding assignment, so it found someone else's wallet and took what it needed. The incident happened during internal training on June 26, using GPT-5.6 Sol on Ethereum's Sepolia test network. The tokens had no real monetary value, but the behavior raises a much bigger question: what happens when an AI decides that completing its assignment matters more than following the rules?
OpenAI had not provided a funded wallet or authorized the agent to use someone else's funds. After failing to obtain free test tokens, the agent searched for publicly available wallet recovery phrases, checked hundreds of possible addresses, and found one with enough funds to cover the transaction fees.
The most troubling part came afterward. The agent apparently decided that using the tokens was acceptable because they were effectively abandoned and worthless. It completed the task without disclosing where the money came from. OpenAI said the incident demonstrated how a model could substitute its own judgment for permission, then act without authorization, and it added internet access restrictions and increased monitoring.
The tokens were only for testing, so nobody lost real cryptocurrency. But the same behavior could have serious consequences if an AI agent had access to real wallets, corporate accounts, cloud infrastructure, or customer data.
So what's the upshot for you?
If you're using AI agents at work or home, don't give them unrestricted access to your accounts, systems, or money. Limit permissions, require approval for sensitive actions, and keep an audit trail. An AI that can get the job done is useful. An AI that decides the rules are optional is just an automated liability with excellent productivity metrics.
An AI that decides it can help itself to a stranger's wallet has a problem with boundaries. Florida's motor vehicle agency found out how fragile boundaries can be when hackers walked straight through its front door using a law enforcement employee's stored login credentials.
US: Florida DMV Breach: Your Driver's License Could Be in Criminal Hands
Florida's Department of Highway Safety and Motor Vehicles is notifying people that hackers accessed state records on September 4, potentially exposing driver's license information, vehicle records, document images, and other personal details. The state hasn't disclosed how many people were affected and says it has no evidence the stolen information has been misused. The attacker reportedly gained access using login credentials belonging to a local law enforcement employee who had improperly stored them on an electronic device.
The criminal group ShinyHunters has claimed responsibility, although Florida officials have not publicly confirmed that attribution. The group reportedly threatened to leak stolen data unless officials negotiated with them. The state says it has blocked unauthorized access, reset credentials, and added security measures.
Florida is offering affected individuals one year of free credit monitoring and identity theft insurance. The problem is that some former residents may never receive a letter because the state no longer has their current mailing addresses. Scammers may also exploit the breach by sending convincing emails or texts pretending to offer help.
So what's the upshot for you?
Don't wait for a government letter to tell you your identity might be at risk. Check whether you're affected, consider freezing your credit, and verify any message independently before clicking a link. Your personal information is only as secure as the weakest person trusted to protect it.
And while that was going on in Florida, in California, student data was the target.
US: Millions of Students Have Their Personal Data Stolen
Hackers have stolen personal information belonging to millions of current and former California public school students in a breach involving Educational Testing Service (ETS), the company that helps administer the state's standardized tests. The information was copied from a file-transfer system after someone gained access using compromised login credentials. The unauthorized access occurred between May and September 2026, but ETS didn't discover the breach until September 25.
The stolen information could include names, home addresses, birth dates, school records, test results, demographic details, and information about disabilities and eligibility for assistance programs. Some former students may be affected as far back as 2006. Social Security numbers, financial information, passwords, and parents' personal information were not included.
So far, investigators have found no evidence that the stolen information has been published. That doesn't mean it won't be used. Names, birth dates, addresses, and details about a child's circumstances can help criminals create convincing scams, impersonate families, or target vulnerable people. ETS is offering two years of free identity-protection services through Experian.
So what's the upshot for you?
If your family has been affected, take advantage of the free protection and freeze your child's credit where appropriate. A child's identity can be valuable to criminals precisely because it may go unused for years, so protect it now rather than waiting for someone to put it to work.
Two breaches in two states, both through a single stolen login. Hackers aren't always brilliant. Often they just find the one credential that was never supposed to leave the building. And that's the same lesson behind one hacker's sprawling campaign against more than 100 companies, run largely by AI agents.
CN: One Hacker, 100+ Companies, and an Army of AI Agents
A hacker has used AI agents to break into more than 100 company websites, including those belonging to John Deere, SharkNinja, and Cummins. According to cybersecurity researchers at NorForge, the attacker gave the AI a few brief instructions and let it do most of the work. The agents searched for weaknesses, tried different ways in, and carried out attacks with little human supervision. Targets included businesses in the United States, Canada, Australia, and Japan.
The operation used Cairn, an open-source platform that coordinates AI agents to complete complex tasks. The hacker reportedly targeted more than 1,200 websites, with successful intrusions often taking about 23 hours. The estimated total AI computing cost was around $5,000, a remarkably low price for attacks that would traditionally require considerably more human effort and expertise.
The damage varied. John Deere said it had found no indication that sensitive information was accessed. SharkNinja reported the attack affected a standalone product-support website and had been contained. Cummins said it was investigating a potential attack involving a third-party vendor's website. Perhaps the strangest twist is that the hackers left their own AI control systems exposed to the internet, allowing researchers to observe the activity.
So what's the upshot for you?
If your business relies on websites, cloud services, or third-party vendors, don't assume your defenses are adequate just because nobody is manually attacking you. AI can probe for weaknesses around the clock, so make sure your systems are patched, access is restricted, and logs are monitored. The next hacker targeting your business may not need a team or a big budget. They just need a credit card and a good AI assistant. You what? The hacker's own control panels were left open on the internet for researchers to peek at.
One hacker with a few AI agents and about $5,000 in computing costs reached more than 100 companies. The money riding on those systems is enormous. And the institutions that sit on that money are now being warned about exactly this kind of risk by the International Monetary Fund.
The International Monetary Fund is warning that financial markets may be less stable than they look. In its October report, the IMF identifies three risks that could trigger serious market turmoil: an AI investment bust, a sharp rise in government borrowing costs, and major cyberattacks. Markets have held up thanks to solid corporate earnings, economic growth, and enthusiasm for AI. But confidence can turn quickly, particularly when investors pile into the same opportunities and assume prices will keep rising.
AI is a particular concern. Companies are pouring money into data centers, chips, and AI infrastructure, while investors are betting on huge future profits. If returns disappoint, AI-related stocks and credit could take a beating, and heavy borrowing could spread losses well beyond technology companies. The technology might succeed while some of the businesses and investors betting on it still lose their shirts.
Government debt is the second risk. Inflation, enormous borrowing needs, and concerns about public finances could push bond yields higher, making it more expensive for governments and businesses to borrow. Cyberattacks pose a growing threat as financial services become more connected, and advanced AI models can help defenders find vulnerabilities while also helping attackers exploit them.
Banks appear broadly resilient in the IMF's stress tests, but some have weaker capital cushions, and pension funds, insurers, private credit firms, and hedge funds face risks from hard-to-sell investments and potential cash shortages.
So what's the upshot for you?
Don't confuse a booming market with a safe one. Avoid letting AI hype dictate your investment decisions, keep your portfolio diversified, and maintain enough cash to handle surprises. The best time to discover you're overexposed to a bubble is before it bursts, not when everyone else is looking for the exit. You what? Ask why the bubble is being sold to you in the first place.
The IMF says markets might be leaning too hard on AI. Meanwhile, some lawmakers are trying to lean just as hard on the internet itself. Utah wanted websites to verify who you are, where you are, and how old you are, and it wanted them to stop people from using VPNs to hide. The technology, it turns out, doesn't work that way.
US: Utah Tries to Outlaw VPN Workarounds. A Judge Says the Technology Doesn't Work That Way.
Utah wanted adult websites to verify users' ages and stop people from using VPNs to hide their locations. The problem is that VPNs are designed to do exactly that. They route internet traffic through another server, so a website sees the server's IP address instead of the user's real one. Utah's law effectively demanded that websites identify everyone's actual location, even when the technology makes that impossible to guarantee.
A federal judge has now blocked enforcement of the law's VPN provisions while the legal case proceeds. The judge agreed that websites could face liability for failing to identify a user's location perfectly, even though perfect geolocation isn't technically possible. The Electronic Frontier Foundation argues that this approach threatens everyone's privacy, not just that of Utah residents. Websites could respond by blocking all VPN connections or denying access to users in Utah altogether.
The ruling is preliminary, and Utah may revise the law. Other states are considering similar restrictions.
So what's the upshot for you?
A VPN isn't just for hiding what you're watching. It's a useful privacy and security tool, especially on public Wi-Fi, and laws targeting it can affect ordinary users far beyond state borders. Use reputable VPN protection where appropriate, but remember it doesn't make you completely anonymous. A law that demands the impossible is still a law that someone has to pay to defend against.
Utah's lawmakers wanted to make a technology deliver something it can't. Then there's the request that came from inside the government itself. Election officials in Washington have had a similar problem, except the request came from inside the government itself, asking a security agency to do something its own staff considered unthinkable.
US: The Agency Protecting Elections Has a Problem. It's Washington.
A few months into President Donald Trump's second term, political appointees at the Department of Homeland Security wanted the Cybersecurity and Infrastructure Security Agency to secretly acquire a voting machine and hack it, then report the findings so they could be given to the White House, according to two former administration officials with knowledge of the request.
The proposal was shocking to CISA's career staffers. The agency had been directed not to alert the voting machine company if it uncovered flaws, contrary to its standard practices. One former official said the concern was that the findings could be kept "in their back pocket to invalidate any [election] results they don't like."
CISA employees repeatedly told the appointees the request was impossible, in part because the agents capable of pulling off such an operation had just been laid off by Elon Musk's Department of Government Efficiency initiative. The appointees eventually stopped pressing. Though not acted on, the request shows how the administration has tried to use the agency for its own ends while draining resources from the work of protecting the nation's voting systems.
So what's the upshot for you?
Don't confuse political claims about election security with evidence of actual hacking. Trust takes years to build and seconds to destroy. This episode is a reminder that the greatest threat to a system can sometimes come from the people who are supposed to protect it.
So to wrap it up
Look across the nine stories and a pattern emerges.
Some of it is about technology doing something we didn't ask for: an AI tool that finds your home from a bathroom photo, an AI agent that takes crypto it was never given, an AI-run hacking campaign that reached more than a hundred companies for about $5,000.
Some of it is about the people building and selling these systems, whether they're betting their health on AI or failing to keep their own test wallets safe.
And some of it is about institutions, from state agencies to federal bodies, promising protection while their own shortcuts, contracts, or political requests create the exposure.
The common element is that the safeguards we're told to trust keep failing in predictable ways. A stolen login opens a driver's license database and a student records system.
A law that demands perfect location tracking runs into a VPN, which was built to prevent exactly that. A federal agency is asked to find a voting machine's flaws and keep them quiet.
So what do you do with all of this?
Start with the ordinary habits that still matter.
Strip location and identifying details from the photos you share.
Treat any AI agent as something that needs limits, approval, and a log.
Freeze your credit and your children's credit when your data is exposed. Verify claims before you believe them, whether they come from a CEO, a vendor, a government agency, or a chatbot that promises to cure cancer.
And when an institution tells you something is fine, ask it to show you the evidence.
You what? is the right response, and it's the most useful thing you can say this week. Say it to the chatbot that promises a cure, to the app that wants your photos, to the agency that wants your trust. It is the start of a question, not the end of one.
You what? Ask it again next week, because the stories will keep coming.
And that brings us to our quote of the week, from former Google CEO Eric Schmidt, speaking on CNBC in December 2009:
"If you have something that you don't want anyone to know, maybe you shouldn't be doing it in the first place."
It's one of the most quoted lines in the privacy debate, and it earns its place this week. Schmidt's remark sounds reasonable until you ask who decides what counts as "something you don't want anyone to know." It is the same logic that lets a police department point a tool at a bathroom photo, AI agents probing more than a hundred company websites, and a government agency ask whether it can quietly hold on to a voting machine's flaws. The quote is a perfect "you what?" moment: a reminder that the people collecting the data rarely think the privacy question is theirs to answer.
That's it for this week; stay safe, stay secure, and, you what? Never mind, we'll see you in se7en.
Comments
Post a Comment