September 22, 2026 — Amazon Locks Meta's Muse Out While Its Zero-Day Waits

Host0:45

Try to check out on Amazon right now using Meta's new AI shopping assistant, and here's what you get. A message saying continued access by an unauthorized AI agent violates Amazon's Conditions of Use. That's Amazon, in plain language, kicking Meta's Muse off its site. Muse launched September eighth. It's Meta's answer to a personal AI agent, one that can browse, fill out forms, and buy things for you while you're doing something else. Meta says Muse has no visibility into your passwords or your payment methods, and that it keeps a full audit trail of everything it does. Amazon isn't buying the part where Muse gets to shop on its site without permission. A spokesperson told reporters that agentic apps like Muse have the same obligations as any other third party, and that Amazon has asked Meta to pull Amazon out of the experience entirely. Here's the twist that makes Amazon's caution look more reasonable. Security researcher Patrick Wardle found a local zero-day in the Mac version of Muse. An unprivileged program already sitting on your computer can quietly change an undocumented setting, redirect where Muse sends its transcription data, and grab the authentication token that controls your account. Wardle put it bluntly. Instead of writing comprehensive Mac malware, an attacker can just leverage the AI assistant itself. So is Amazon protecting shoppers, or protecting its own turf from a fast-growing rival? Muse pulled in one point eight million downloads in the U.S. and Canada in its first twelve days, ahead of ChatGPT's early pace. Probably both things are true at once. The open question is whether Meta patches this before that zero-day gets used for something worse than a demo. None of that has anything to do with what grounded flights across the Northeast yesterday, but it's a good reminder that most tech failures start small and boring.

2:51

Picture six hundred feet of fiber optic cable. That's the whole story. A construction crew cut it by accident on September twenty first, and it happened to run through the facility outside Philadelphia that handles radar approach control for a huge chunk of Northeast air traffic. The primary circuit at that facility was already old and due for replacement. FAA Administrator Bryan Bedford told CBS News that when the backup fiber got cut too, the entire communications setup at that facility went down, and the FAA had no choice but to slow or stop traffic. The damage spread fast. Newark saw more than five hundred cancellations and over two hundred delays. LaGuardia and JFK had dozens of cancellations and hundreds of delays. Philadelphia logged around one hundred eighty cancellations and nearly four hundred delays. Transportation Secretary Sean Duffy confirmed a construction crew accidentally cut into the line, though who was actually responsible took most of the day to sort out. Amtrak got blamed first, then it turned out the work was happening on New Jersey Transit property, with Verizon confirming it was their fiber, just not their crew that cut it. Lines got fixed and flights resumed by evening. But this is one cut cable grounding a huge piece of the country's air travel. The real question isn't whether they'll fix this line. It's how many other single points of failure like this one are sitting under Northeast infrastructure right now. Infrastructure failures are physical. The next one is entirely about data, and it's going to cost Google a lot more than a bad day of flight delays.

4:39

Four hundred three million euros. That's the fine Ireland's data protection regulator just handed Google, and it's not about anything flashy like a chatbot. It's about location data. The regulator looked at three features. Web and App Activity, Location History, and Location Accuracy, covering processing from May of twenty eighteen through February of twenty twenty. Investigators found Google broke GDPR's basic rules on lawfulness and fairness for two of those features, failed on transparency and accountability for the third, and held onto data longer than it should have. Deputy Commissioner Graham Doyle put it simply. Location data gets a high level of protection under European law, and processing it has to be lawful, fair, and transparent. Google now has six months to bring its systems into compliance. Google hasn't said publicly whether it will appeal. That's really the open question. Does this ruling just apply to old settings from twenty eighteen, or does it set a precedent for how any company that infers your location from multiple signals has to disclose that going forward. If it's the second one, this fine is the cheap part. Ireland is going after how Google handles your location data. California just went after something bigger, the actual physical footprint of the machines behind it.

6:08

California signed seven bills aimed straight at data centers, the buildings full of servers that make the cloud and AI actually run. The bills tighten oversight on energy use, water use, and local permitting. The practical effect, according to reporting, is that operators will have to absorb more of the utility and environmental costs locally instead of pushing them onto ratepayers or getting easy approvals. That changes the math on where a company decides to build its next big server farm. Data center operators and their trade groups are going to argue this raises costs and just pushes new construction to states with looser rules. Legislators are going to argue that's the point, that communities shouldn't subsidize the water and power bill for somebody else's AI training run. Nobody knows yet how much this actually redirects investment away from California. But it lands right when demand for that power is about to go up even more, because the biggest cloud companies are now racing to lock down their own electricity supply. Google is one of those companies, and its answer to the power problem is basically to help build its own power plants.

7:24

Bloomberg reported that Google struck a deal with Georgia Power to boost nuclear capacity, part of a broader pattern of cloud companies funding electricity generation directly instead of just buying it off the grid. Separately, Kairos Power says it's getting up to one hundred million dollars from Samsung Group to build a reactor that reporting describes as meant to supply Google. That's a shift from the old model, where a tech company signs a power purchase agreement and lets someone else worry about generation, to a new one where the tech company helps fund the reactor itself. Worth being careful with the word dedicated here. Reporting describes partner plans and financing commitments, but the actual contractual exclusivity, who gets first claim on the power this reactor produces, isn't public yet. That's the kind of detail that shows up in filings, not press releases. The bigger question is whether this actually speeds up nuclear power coming online at all, or whether it just means Google jumps to the front of a very long permitting line while everyone else waits. That's the slow, expensive end of the AI story. Now let's do the fast, cheap end.

8:40

This is Fast Boot. Four stories, thirty seconds each. Here we go. RedVDS. Microsoft and Google teamed up to take down a cybercrime marketplace called RedVDS, which sold ready-made virtual machines preloaded with malware, an operation reportedly worth sixty six million dollars. It's a real disruption to the criminal supply chain, but marketplaces like this tend to grow back, so call it a pruning, not a removal. Paramount and Warner Bros. Discovery. Paramount settled a state antitrust lawsuit that was standing between it and its one hundred ten billion dollar acquisition of Warner Bros. Discovery. The settlement clears a major roadblock, though it's not clear yet what Paramount gave up to get there. AWS and Nvidia. Reports say AWS plans to buy more than three million additional Nvidia chips before twenty twenty-nine, on top of the thousands it already has. At that scale, somebody's going to joke that Amazon has more GPUs than employees, and honestly, they might not be wrong. OpenAI's math problem, literally. OpenAI formed an advisory group of mathematicians after claiming its model solved the Navier-Stokes problem, one of math's hardest open questions, plus more than one hundred other open problems. Only one of the nine people on that new panel had also signed a letter from twenty five Fields Medal winning mathematicians raising concerns about how AI labs handle math claims. Make of that what you will.

10:18

That last one was about OpenAI's math claims. This next one is about OpenAI and a much heavier kind of claim. The Province of British Columbia is suing OpenAI. Bloomberg reports the lawsuit alleges OpenAI failed to flag a shooter's ChatGPT activity to police before a mass shooting. The legal argument, as reported, is that OpenAI had signals in that person's conversations pointing to planned violence and didn't notify anyone. That's a heavy claim, and right now it's just that, a claim in a filed lawsuit that OpenAI hasn't had its full say on yet. If a court agrees with British Columbia, the ripple effect goes well past OpenAI. It would mean AI companies have some kind of legal duty to monitor conversations and proactively report suspected violence to authorities. Civil liberties groups are going to push back hard on that, because it starts to look a lot like mandated surveillance of private conversations. OpenAI has not yet had its full say. The question sitting under this whole case is where the line falls between a company that just built a text box, and a company that's now expected to watch what people type into it.

11:36

Today, Amazon blocked Meta's Muse from shopping while a security researcher found a hole in its Mac app, a fiber cut grounded flights across the Northeast, Ireland fined Google four hundred three million euros over location data, California passed seven new data center bills, Google helped fund new nuclear power capacity, Microsoft and Google took down a cybercrime marketplace, Paramount cleared a path for its Warner Bros. Discovery deal, AWS is buying millions more Nvidia chips, OpenAI formed a math advisory group, and British Columbia sued OpenAI over a mass shooting. Full sources for every story are in the show notes on our website. If this show is useful to you, follow The Daily Reboot wherever you're listening, so tomorrow's episode just shows up. That's the show for today. Talk to you tomorrow.