Sam: The AI's final circuit board design? It called for a resistor they stopped making in 1982.
Kai: I'm Kai.
Sam: And I'm Sam.
Kai: And this is Code Watch for Friday, September 5th, 2026, where we track the most-starred, most-forked, and most-talked-about projects in the dev world.
Sam: And we have three huge stories on deck.
Kai: We're starting with the big question: can AI actually design our circuit boards yet? The answer... is not what you think.
Sam: Then, there's an actively exploited sandbox escape in all versions of Chromium. And yes, we do mean all of them.
Kai: And finally, an AI model from Anthropic just tackled a 358-year-old math problem.
Kai: Alright, let's dive into hardware. A new post on EEBench asks the question on every hardware hacker's mind: 'Can AI design circuit boards yet?'
Sam: And based on my intro, you can probably guess the answer is... complicated.
Kai: Come on, Sam! This is the holy grail. You feed an AI a schematic, a set of constraints, and it spits out a production-ready PCB layout. The project they tested, something called 'PCB-GPT', routed a pretty complex board in under a minute.
Sam: It routed the board, sure. But as I mentioned, it also used components that are four decades out of production.
Kai: Okay, a minor detail!
Sam: A minor detail? Kai, it's like asking an architect for a skyscraper and they design it using asbestos and lead paint. The AI also, in one test, completely forgot to connect the board to a power source.
Kai: [laughs] Okay, okay, that's... not ideal. But it's a stepping stone! It's autocomplete for hardware engineers. It gets you 90% of the way there.
Sam: Sure, if you're an expert who can spot that missing 10%—which includes errors that turn the whole thing into a paperweight. That's not a time-saver, that's just a new kind of debugging hell.
Kai: But think of it as a force multiplier! For makers and small teams, this means you can iterate on designs way faster. So what if the first draft is weird? You fix it and get a new one in 60 seconds.
Sam: No, what this really means is that if you're not a hardware engineer, you can't trust these tools. Period. Don't just fire up an AI and send the file straight to the fab house.
Kai: So the answer to 'Can AI design circuit boards?' is 'Yes, but badly.' [laughs] For now.
Sam: [sighs] Precisely. Okay, let's move on to something much more immediate and much more serious: CVE-2026-85046.
Kai: That's a mouthful. What is it?
Sam: It's a remote code execution vulnerability—a sandbox escape—in the V8 JavaScript engine. That's the engine in Chromium. Plain English? A malicious website could potentially take over your entire computer.
Kai: Whoa. And this is in Chromium, so that's not just Chrome, right? That's Edge, Brave, Opera, Electron apps...
Sam: Everything. And the NIST report confirms it's being 'actively exploited in the wild.' This isn't a theoretical problem. The attackers are already using it.
Kai: Whoa. Okay, so what do people do? Just how bad is this?
Sam: It's as bad as it gets for a browser vulnerability. And what you do is simple. As soon as you finish this podcast—no, right now, pause us—go to your browser settings and hit 'Update.'
Kai: I'm literally checking for an update on my machine as we speak.
Sam: Good. For once, your early-adopter impulse is the right one. The takeaway for everyone listening is non-negotiable: update your browsers. All of them. And check your Electron apps for updates, too. This is the real deal.
Kai: [relieved] Okay, update is downloading. Alright. Let's shift gears from the terrifying to the... deeply mathematical. Anthropic just published a paper on formalizing Fermat's Last Theorem.
Sam: You mean the theorem that says while a-squared plus b-squared can equal c-squared, there are no solutions for cubes, or any higher power. The one Andrew Wiles proved back in the 90s.
Kai: The very same! But Wiles's proof is over 100 pages of dense, complex human reasoning. What Anthropic did was train a model to translate that whole proof into a machine-readable, formally-verified language called Lean.
Sam: Okay, but hold on. The AI didn't solve the theorem. It didn't create the proof. It was a translator.
Kai: But that's what's so amazing! This is a massive translation task that no one had fully cracked. It's not translating words, it's translating logical concepts. The model even found and fixed bugs in existing human-written formalizations.
Sam: Which is impressive as a feat of engineering, I'll grant you. It's an incredibly powerful spell-checker for mathematicians, but it's not the same as creative discovery. It's not the spark of genius Wiles had.
Kai: I think you're underselling what this unlocks. Now we can start to formally verify all sorts of complex systems. Imagine verifying a new crypto algorithm, or the safety controls for a rocket.
Sam: And I agree, that is hugely valuable. It reduces the chance of human error. But we should be clear: this is an AI assistant, a powerful tool for experts, not a replacement for them. The headline isn't 'AI Solves Fermat's Last Theorem.'
Kai: Fine, the headline is 'AI Helps Prove Fermat's Last Theorem Is Correctly Proven.' Happy?
Sam: [chuckles] Much. And that's a pretty good place to be.
Kai: Alright, so to recap: AI circuit board designers are coming along, but they still need a human babysitter to make sure they connect the power.
Sam: Next, a critical RCE in Chromium means you need to update all your browsers. Right now. I'm not kidding.
Kai: And an AI just proved it's the world's most diligent math grad student by formalizing the proof of Fermat's Last Theorem.
Sam: A solid day of updates.
Kai: And before we go, a quick nod to history. On this day, September 5th, 1977, NASA launched Voyager 1.
Sam: Wow. Still running, 49 years later. It's on custom hardware and software with less computing power than the chip in my coffee machine. And it has an 8-track tape recorder for data storage.
Kai: [laughs] It's the ultimate legacy system. We're talking about AI-designed circuits and it's out there still working on 70s tech. There's a lesson in there somewhere.
Kai: That's our show for Friday, September 5th. We'll be back tomorrow with more from the world of open source.
Sam: We will. Now go update your browsers. Seriously.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.