Voters Reject AI Data Centers, Google's Privacy Pivot, and DigiCert Exposed | Tech News of the Week

[00:00:03.01] - Ned
I think we're all special this, this week.

[00:00:06.23] - Chris
We are all special on this special day? No.

[00:00:11.02] - Ned
On this very special episode of Chaos Lever. Oh dear.

[00:00:15.04] - Chris
Well, all the episodes are special, that's for sure.

[00:00:18.03] - Ned
In their own unique ways, yes. Oh, there were so many special episodes in the '80s and '90s. For various sitcoms.

[00:00:32.05] - Chris
Well, it was just like a rule. They were like, well, things have been funny for 20 straight weeks. Now we have to rip your heart out and stomp on it.

[00:00:38.08] - Ned
Yes. Or just make you feel deeply uncomfortable about topics that you are not prepared to deal with.

[00:00:44.06] - Chris
That's true.

[00:00:46.18] - Ned
Yeah.

[00:00:47.23] - Chris
But I mean, if we remember, Bambi was rated G, and that started with parent murder.

[00:00:54.02] - Ned
As all good Disney movies do. That really is like the, the first like beat of any Disney plot is kill one of the parents.

[00:01:06.04] - Chris
Someone precious dies and then we sing about it.

[00:01:10.07] - Ned
Hello, alleged human, and welcome to the Chaos Lever Podcast. My name is Ned and I'm definitely not a robot. I'm a real human person with parents that I love dearly. It's getting dark. With me is Chris, who's also here and has parents. Maybe.

[00:01:29.09] - Chris
I can neither confirm nor deny.

[00:01:31.06] - Ned
That is about the level of response that I would expect from you.

[00:01:35.20] - Chris
My best understanding is that I was genetically created in a lab at the age of 35.

[00:01:42.05] - Ned
35? I think you're setting the age a little low on that one.

[00:01:47.03] - Chris
Well, that was 70 years ago.

[00:01:49.15] - Ned
Valid point. Which explains your obsession with the 1950s.

[00:01:56.06] - Chris
I mean, it was the era of doo-wop.

[00:01:59.06] - Ned
Doo-wop, doo-wop, doo-wop, bop, bop.

[00:02:02.00] - Chris
Thank you.

[00:02:02.20] - Ned
Let's talk about some tech garbage. Certification company DigiCert certified insecure. Try saying that 5 times fast.

[00:02:14.15] - Chris
I will not.

[00:02:15.07] - Ned
All right. Moving right along. DigiCert is one of the best known certificate authorities on the web. If you peek into the trusted root certs on your browser, DigiCert will certainly be among them. So while your browser may trust DigiCert's chain to secure your data, maybe you shouldn't. Back in February of 2025, DigiCert accidentally placed one of their internal business systems on a virtual network that was publicly accessible from the internet. That's, um, it's not, it's not great.

[00:02:52.19] - Chris
Less than ideal. Less than ideal.

[00:02:55.13] - Ned
So what was on this system? Uh, it was an archive system that held historical data about customers, including their names, addresses, emails, And in some cases, photo IDs and passports.

[00:03:10.11] - Chris
Aye, aye, aye.

[00:03:11.18] - Ned
Now you might rightly wonder, why would DigiCert have someone's passport on file? Seems odd.

[00:03:19.21] - Chris
I do wonder that.

[00:03:20.23] - Ned
Well, I know that we all use Let's Encrypt and ACME to issue our TLS certs these days, but there was a time when getting a certificate for your website was a bit more of a hassle. When I had to get certificates for a small business back in 2008, part of that process was proving 3 things. First, that the company owned the domain in question, that I was an authorized representative for that company, and that I was who I claimed to be. The process of verification took over a week and required the submission of many documents. Signed affidavits, promises made under a full moon with Mercury rising, and a lock of my firstborn's hair. I'm only slightly overstating this. And that was only if you wanted to get the regular certificate. If you wanted the trusted super enterprise certificate, you had to pay $5,000 more and also make a blood sacrifice or something. So yeah. The idea that DigiCert would have your passport photo and ID on file is actually not all that surprising. What is surprising is that this internal archive system sat on the open internet for almost 16 months. You heard that right.

[00:04:43.21] - Ned
Server sat there from February 2025 until July 2026. And because the archive used the same account platform as the customer-facing systems, You could self-register for a DigiCert account and use that account to access the archive. Despite finding the service and fixing the networking, DigiCert failed to realize the severity of the issue until October of this year, when they began to investigate in earnest. DigiCert has begun sending messages to potentially affected parties and fixed many of the security issues that led to this oversight. You can check out their full postmortem at the link in the show notes. Still no word about getting back my son's lock of hair though.

[00:05:33.02] - Chris
Data centers have been weighed, they have been measured, and they have been found wanting. This past Tuesday, America, you may have heard of it, had a pretty consequential election.

[00:05:48.03] - Ned
Yeah.

[00:05:49.16] - Chris
Now, this is not a political podcast, so we will skip the headlines of those election results, but there was something down ballot that definitely deserves our attention because across 10 states, 41 various local measures were asking some sort of question about whether or not new data centers were a good idea. The wording varied, but that's the gist of it. And in 40 of those 41, The answer was no.

[00:06:20.09] - Ned
Oh,

[00:06:21.06] - Chris
40 out of 41 nos. That is 97.5%. If you round that up, that's like 400%. Talk about a referendum. Yeah. AI has been the author of a lot of controversies over the past few years, and data centers that house that AI have increasingly become a very important local issue. issue. Now, a coworker of mine said about data centers, they've been around for ages. People have probably driven past 100 of them in their lifetimes and never even noticed.

[00:06:55.03] - Ned
Yep.

[00:06:55.16] - Chris
And that's a true statement for old-style colocation-based data centers. Those data centers were built carefully over time with proper planning, insulation, water inflow and outflow, and power considerations. They worked with the grid, on the grid. They did not destroy the grid.

[00:07:17.08] - Ned
Yes.

[00:07:19.16] - Chris
AI data centers, on the other hand, famously are not that. They are built fast and cheap, usually under the COVID of night. They are most likely not up to code. They keep the neighbors up at night and siphon off the local water. Witness 1 for the prosecution, The Twitter AI data center gas-powered turbines that were installed without permits and left for 200 days that we talked about a few weeks ago.

[00:07:46.10] - Ned
Yeah.

[00:07:47.08] - Chris
And these AI data centers usually require public funding, or at least public approval to even happen, let alone the crazy tax breaks and/or some payments that are made under the table, allegedly.

[00:08:00.06] - Ned
Clear bribery, I think, is what it's called.

[00:08:03.22] - Chris
This vote Showed pretty clearly what the voting public thinks about this. What do you want to call it? Phenomenon?

[00:08:11.07] - Ned
Racket.

[00:08:12.16] - Chris
Oh, you're getting good words today. To pick just one example, Kansas, of What's the Matter With fame, voted to ban new, quote, high-impact data centers for 10 years. Just no. For 10 fucking years.

[00:08:32.17] - Ned
Wow.

[00:08:33.14] - Chris
Now admittedly, this was limited to the city of El Dorado, which is in Butler County, Kansas. But still, I mean, damn.

[00:08:40.23] - Ned
Damn.

[00:08:42.03] - Chris
Emporia, Kansas did something similar, if that helps, which I think it does. After the election and this just avalanche of a result, OpenAI pledged to pay full freight for their data centers Solicit community input about new projects and do what they could to protect ratepayers when it comes to utilities. You know, I mean, all the shit they could have been doing this entire time and thus never ended up in this position in the first fucking place. Yes. Their own chief global affairs officer said that the blueprint for this new data center model has been in the works, quote, since August, and that publishing it now, quote, helps put more energy into the momentum before the new Congress sits. Aha.

[00:09:35.21] - Ned
Sure, Jan.

[00:09:37.18] - Chris
I'm not saying that OpenAI is insincere.

[00:09:40.18] - Ned
Yes, you are.

[00:09:45.19] - Chris
I'm just saying that I've never once seen a company discover revolutionary operational moral principles just on a random fucking Thursday. Point is, the public had their chance to tell their AI overlords exactly what they think about these data centers. And there is at least a fig leaf of a chance that at least one of them, or I guess if we're being honest, their PR department actually listened.

[00:10:17.06] - Ned
GVisor is not a horribly outdated clothing line from Lids. Now, I couldn't remember if Lids was spelled with a Z because I don't go to the mall. So I used an S, but you know, this being an audio podcast, you care about that. Anyway, last week, Googley— Google— Googley?

[00:10:37.14] - Chris
Googley.

[00:10:38.09] - Chefs
Great googly moogly.

[00:10:40.13] - Ned
Officially donated 3 projects to the CNCF that all have to do with AI agent things. The 3 projects in question are gVisor, Agent Substrate, and OpenRL. As noted already, gVisor is not a Kangol knockoff for dude bros.

[00:11:00.03] - Chris
You definitely spelled Kangol wrong.

[00:11:01.16] - Ned
Oh yeah. Instead, you can think of gVisor as an isolation technology for containers. gVisor provides a lightweight kernel in user space that intercepts system calls from a container app And filters those calls before passing them along to the actual kernel. It's similar in nature to the hypervisor layer that protects the host operating system from the guest operating systems that virtual machines run. gVisor is especially useful for systems that will be running agentic workloads. When you have a traditional container app, you have a pretty good idea of what that app Is going to try to do. After all, the code is deterministic. With agents, they will try all manner of fuckery to accomplish their task, including controlling the speeds of fans in a system to try and communicate with other agents. Yes, that's a real thing. With gVisor sitting between the operating system and the Container App, that kind of behavior is a lot harder to pull off. Agent Substrate is a way of scaling out the number of agents that a Kubernetes cluster can effectively run. Most agents are idle like 80% of the time. But the pod hosting that agent sits there taking up resources.

[00:12:22.16] - Ned
If you kill the pod, the agent loses its context, which is not good. So Agent Substrate creates a persistent data layer for agents to store their context in, and allows an agent to be put in a dormant state while not actively serving requests. Pods can be packed with multiple agents and use gVisor as an isolation layer, leading to much more efficient resource consumption. The last one, OpenRL, is the project I'm least familiar with, but it seems to be something that uses Kubernetes clusters to perform reinforcement learning And supervised fine-tuning for models. Hooray! Why is Google doing all this? And does it mean that they are abandoning these projects? The main reason for the donation is that some vendors won't embrace a solution while Google is the sole maintainer. Kubernetes itself is a good example. By moving governance and maintenance out of Google's hands and into the CNCF, The company is hoping that these projects will become standards for the industry. The net result remains to be seen, but I think overall this is a good move. Score one for Google. I wonder if this is the beginning of a trend for the ad-serving, privacy-stealing giant.

[00:13:44.23] - Chris
Company you already knew was terrible proudly, if quietly, continues to get worse. From the this is awful from a few weeks ago, I missed it the first time around, and I blame Ned specifically department, advertising company Google decided to update their privacy policy again, secretly in the dead of night under a cloudy sky during a new moon.

[00:14:12.16] - Ned
Near Halloween.

[00:14:15.09] - Chris
In short, they have decided to modify a kind of important sentence that used to live in the ads personalized section of their privacy policy. Prior to October 16th, it read, quote, we don't show you personalized ads based on your content from Drive, Gmail, or Photos, unquote. That sentence now reads, quote, we don't show you personalized ads based on your content from Drive, Gmail, or Photos unless you choose to make data from these services available to other Google services that support personalized ads, unquote. Quote,

[00:14:53.16] - Ned
Gemini.

[00:14:56.01] - Chris
It could be anything.

[00:14:57.21] - Ned
Gemini.

[00:14:58.09] - Chris
But it's totally Gemini. Yeah,

[00:14:59.14] - Ned
it's totally Gemini.

[00:15:02.07] - Chris
The connections to other apps is still, for the time being at least, entirely opt-in. So why did Google feel the need to modify that sentence at all, you might ask? Well, way back when they deleted Don't Be Evil, and one can only assume they did that because their shareholders thought it would be best to leave the door open to being evil later. Fast forward to now.

[00:15:26.01] - Ned
Yeah.

[00:15:27.16] - Chris
The Ars Technica article that I'm using for this works out the logic. Sure, I mean, your data is safe all by itself, but if you use Google AI— oh, and by the way, you more or less have to use it in order to use Google Search— guess what? Google AI serves personalized ads. So the pipe runs like such, like so, like something. Your email goes into Google AI. Google AI produces summaries and insights. AI mode uses those to answer questions, and the ads shown alongside that answer are targeted to you using your personalized information. Bob's your uncle. Your high school sweetheart's name was Jennifer. Your Social Security number is 123456789. By some goddamn degree, you lonely, sweaty swamp monster.

[00:16:22.09] - Ned
Hey.

[00:16:23.14] - Chris
Back to that opt-in part. Google says extremely quietly, surrounded by their lawyers, you did opt in. But ours, quote, Google AI uses various prompts and pop-ups to entice users to turn it on without fully explaining the consequences. You may have turned it on without realizing. For example, would you like better answers? Google asks innocently.

[00:16:52.12] - Ned
Hmm.

[00:16:54.01] - Chris
Which, if we all know our good friend the Googlenator, is exactly what they're hoping you'll say yes to.

[00:17:00.23] - Ned
Only have to say yes once. GitHub is crossing the Rubicon to embrace Azure Blobs. I swear This sentence will make sense eventually. Unless you've been living under a rock, which now that I think about it might not be a bad idea, you're probably aware of GitHub's recent reliability woes. The company's own reported uptime took some serious hits in 2026, including a partial outage back in March that basically created a developer snow day. It was fun. We went sledding. As part of their annual conference, GitHub Universe, the code storage company has posted a series of blogs about what went wrong, why it went wrong, and what they've done to fix it. The why it went wrong is pretty simple. AI. Of course it's fucking AI. Agentic coding has led to a massive increase in code activity, for GitHub-hosted repositories. And it's not just the velocity of code changes, but also the sheer volume. I know people say that I like to ramble, but jumping Jesus on a pogo stick, have you asked an LLM to write anything? I've gotten back 500-word essays that could have been a text message and stayed under the 144-character limit. ChatGPT and its ilk are famously verbose, and it turns out that the same thing is applied to how they write code.

[00:18:42.02] - Ned
To quote the post, quote, between September 2025 and August 2026, it increased— that is, the amount of traffic increased to more than 2x its previous level, from 218 billion events per month To 473 billion events per month. In September alone, developers and agents made 7.38 billion commits on GitHub, more than 5 times as many a year earlier. So not only is it increasing, it's increasing month by month on almost an exponential scale. And it is this absolute tsunami of writes that is tripping up GitHub. Their original architecture scaled compute and storage together, meaning to scale up, they relied on adding replicas that participated in a quorum. As the quorum grew, writes became slower, and the loss of a replica would have a major negative impact on reads and writes. The agentic overwhelm also hit GitHub while it was trying to migrate to Microsoft Azure, which was having its own capacity problems. The solution was a new architecture they're calling Rubicon. This separates compute from storage and reads from writes. The blog post goes into some serious detail about how a read or write is processed for a repository, and I think it's really interesting reading if you're curious about scaling distributed systems.

[00:20:17.19] - Ned
To boil it all down, the team now creates multiple virtual replicas each repository. The replicas can service reads or writes and scale independently of each other. A separate service maintains a reference for all packs in the repository, which the virtual replicas can query to see what's missing in their local cache. The whole thing is backed by Azure Blob Storage, which holds packs, indexes, records, leases, and read accelerators. When planning the new approach, The GitHub team didn't just plan to handle the current load, they were designing for a system that could survive a further increase of 2 orders of magnitude. They're assuming this agentic software development thing is unlikely to be going anywhere and that code volume and velocity is only going to increase. The updated platform is now live and has been tested to support 5,000 concurrent agents Sending 7,200 pushes per second to a single repository while that repository is still serving reads with a low latency. That's the sort of thing that would make the old architecture fall over and cry and run home to its mommy. GitHub has been dealing with some substantial challenges and technical debt over the last couple of years.

[00:21:37.14] - Ned
Here's hoping that this newly rolled out platform takes some of the pressure off the SRE and development teams And lets them get back to improving the service instead of just trying to stay afloat.

[00:21:49.14] - Chris
AI slop is coming for Christmas again.

[00:21:54.00] - Ned
Yay.

[00:21:54.14] - Chris
This time in the name of selling $2,000 jeans. Ah, yes, the holiday season.

[00:22:04.20] - Ned
I'm sorry, did I say that out loud?

[00:22:09.09] - Chris
Ah, I think you're saying what we're all thinking. Christmas, which as we all know is the most important holiday in the history of Earth and most other planets. The word Christmas, if you don't know this, translated from the original Aramaic actually means, actually fuck your soul, go out and buy as much overpriced shit as possible. Jesus will thank you, probably. I mean, I know Visa will, but oh fuck, I mean, okay, hallelujah or whatever, buy more shit now.

[00:22:38.10] - Ned
Wow, that's a lot to pack into one word.

[00:22:40.06] - Chris
It's a very expressive language. A few years back, we had what I used to think was the worst holiday commercial to ever run on the national airwaves.

[00:22:52.06] - Ned
Still have nightmares.

[00:22:53.12] - Chris
That was when beverage distribution company Coca-Cola decided to run an AI ad that was incomprehensible and stupid but cheap,

[00:23:03.15] - Ned
Was it?

[00:23:05.17] - Chris
I mean, it was cheaper than doing it right.

[00:23:07.09] - Ned
All right.

[00:23:09.00] - Chris
And the fun thing is Coke still runs it and they keep running it over and over and over. So that was bad. But at least we were content knowing that everyone, and let's be honest here, including Coke, knew that it was bad.

[00:23:25.22] - Ned
Yep.

[00:23:26.14] - Chris
They never admitted it, but I mean, Come on,

[00:23:29.21] - Ned
they got eyes.

[00:23:32.00] - Chris
This year, though, we have a whole new level of bad. And it comes from a company that just— look, I don't, I don't understand why these companies even exist.

[00:23:43.11] - Ned
Fair.

[00:23:44.20] - Chris
I watched The Devil Wears Prada. I learned nothing.

[00:23:49.00] - Ned
Okay.

[00:23:49.22] - Chris
Christian Dior, a company famous for being on the absolute cutting edge of fashion, and by which I mean expensive, stupid, flimsy clothing and other insignificant baubles, decided to put out a long-form ad for Christmas. It was about 6 minutes long, and they went to great lengths to claim that it was hand-animated. They even showed a behind-the-scenes video that had animated stills.

[00:24:19.17] - Ned
Okay.

[00:24:20.20] - Chris
But here's the thing. When you watch this 6-minute ad, it kind of sucks. It's clearly not right. All

[00:24:32.09] - Ned
right,

[00:24:33.18] - Chris
I'm not an animator myself, so I was not able to go into the weeds, but I'm just saying I have seen Scooby-Doo episodes where character movement looked more convincing.

[00:24:44.06] - Ned
Yeah.

[00:24:46.08] - Chris
Now, real animators Have absolutely torn the film apart, showing both nakedly obvious and subtle clues that this was not hand-animated. It was in fact deeply gen AI-based.

[00:25:01.13] - Ned
Awesome.

[00:25:03.08] - Chris
We're talking hands with multiple fingers. We're talking eyeballs in the wrong place. We're talking movement that just doesn't make sense. Link in the show notes. They have screen grabs. It's pretty obvious.

[00:25:16.02] - Ned
All right.

[00:25:17.13] - Chris
The director of the ad is no saint either, if those comments are to be believed. Uh, there are discussions of him being a plagiarist as well as a shitty boss. Now, these are allegations. We are not saying anything out loud. Please don't sue me, France. I only mention them because he has also had a number of people holding him up and stumping for his anti-AI process. And based on what I've seen, uh, to quote a great philosopher, that anti-A process, that was a fucking lie. Oh. Now the problem here is what kind of fallout could actually be meaningful or make any kind of change? There's not going to be any consequences aside from Dior inevitably pulling the ad and issuing some kind of mealy-mouthed non-apology. I mean, it does— this whole thing does a tremendous disservice to real animation, a difficult and dying art that absolutely deserves more eyeballs. Support actual artists, for God's sake.

[00:26:25.12] - Ned
Yeah.

[00:26:27.00] - Chris
Not this shit. This shit sucks, and I'm mad at everything. I need some proper classic animation. Studio Ghibli, for example. I think I'll go watch, oh, I don't know, Grave of the Fireflies. That's sure to make me feel better.

[00:26:42.06] - Ned
Certainly sounds like it from the title. Hey, thanks for listening or something. I guess you found it worthwhile enough if you made it all the way to the end. So congratulations to you, friend. You accomplished something today. There you go. Sit on the couch, watch an AI-generated animation and get extremely mad about it. And then tell us on the internet. You've earned it. You can find more about the show by visiting our website, chaoslever.com, where you'll find show notes, blog posts, and general tomfoolery. And if you wanna sponsor Chaos Lever, stop sending me random emails about your weird AI app. Haven't we made it pretty clear how we feel about AI? Jesus. We'll be back next week to see what fresh hell is upon us. Ta-ta for now.

[00:27:36.00] - Chris
The Ars article— Arsicle?

[00:27:38.23] - Ned
No.