Ivy: So a worm gets into a company's code by hijacking their AI assistant. And the kicker? They didn't even notice for three weeks.
Marcus: I'm Marcus.
Ivy: And I'm Ivy.
Marcus: It's Thursday, September 17th. And this is Broke It, your daily AI news download. We've got a lot to get to today.
Marcus: Coming up: a terrifying new supply chain attack where the hacker's accomplice... was the office AI.
Ivy: Plus, Al Gore is surprisingly chill about AI's carbon footprint. We'll get into why.
Marcus: And a company that builds simulated worlds for AI voices just landed a massive funding round.
Marcus: Okay, let's dig into this new Mandiant report. Ivy, it is... a doozy.
Ivy: One word for it is right. Mandiant reports that an attacker compromised a popular AI coding assistant... they won't say which one, of course.
Marcus: And they didn't just steal data. They used their access to the assistant's backend to push a malicious update. A worm.
Ivy: Which then used the AI's permissions inside developer environments to spread through hundreds of private code repos. It was hunting for secrets—API keys, credentials, all the good stuff.
Marcus: That is both horrifying and brilliant. The AI tool is trusted, it has access everywhere... It's the perfect way in.
Ivy: It's the perfect way in if your security is a joke. How do you not have systems that flag a tool suddenly reading every file it can and shipping data out?
Marcus: Because you trust the tool! It's from a major vendor, it's supposed to be secure—
Ivy: No, you trust nothing. It's called Zero Trust for a reason, Marcus. This is a classic supply chain attack, just with a new, trendy AI entry point.
Marcus: Okay, but the scale of it! And the attacker used another AI to analyze the code it was stealing to find the most valuable credentials first. It's AI-on-AI crime!
Ivy: So what's the lesson for developers or companies using these tools?
Marcus: It means your threat model just changed. That AI assistant isn't a fancy spellchecker. It's a dependency, and you have to treat it like one.
Ivy: It means you need to aggressively log, monitor, and lock down the permissions for these AI tools. Because the vendors clearly aren't doing it for you.
Ivy: Alright, let's switch gears from immediate threats to... existential ones. Al Gore talked to TechCrunch about AI, and his take was not what I expected.
Marcus: Right, you'd think Mr. Climate Change himself would be all over the massive energy consumption of AI data centers. But he kind of... shrugged?
Ivy: He didn't exactly shrug. He said it's 'interesting' that the people building the AI are the same ones 'issuing the most dire warnings' about it. He seems more worried about that.
Marcus: Hold on. So the guy who's been screaming about emissions for 30 years isn't worried about the industry that's about to consume the energy of a small country? That feels... off.
Ivy: I think his point is about triage. We know how to make greener energy and more efficient data centers. That's an engineering problem we can solve. He's less sure we can solve... containing a superintelligence we don't understand.
Marcus: I just... I can't get past the hypocrisy. We can't ignore a definite, massive climate impact today because of a hypothetical, maybe-never risk tomorrow.
Ivy: I see it differently. He's looking past the current outrage cycle—the data centers—and focusing on the risk the actual experts are flagging. It shows he's listening to the technologists, not just the headlines.
Marcus: So what's the takeaway here? Don't worry about the electricity bill because the robot apocalypse is coming?
Ivy: The takeaway is that the conversation is growing up. We're moving past just 'AI is bad for the planet' to a real debate about different kinds of risk. And Al Gore, of all people, is helping lead it.
Marcus: Alright, let's follow the money... all the way to Iceland.
Ivy: A company called Treble just closed an 18-million-dollar Series A.
Marcus: Eighteen million! And what they do is so cool. They don't make the AI voices—they build the worlds those voices live in.
Ivy: Explain. Because 'building the world for voice models' sounds like something from a bad sci-fi movie.
Marcus: It's a platform that simulates acoustics. So if you're building, say, a voice assistant for a new wearable, you can test how it'll sound in a quiet library versus a noisy car or an echoey cathedral, all from your desk.
Ivy: So it's audio middleware. It helps AI companies train and test their voice products for different environments. That's... surprisingly practical.
Marcus: It's huge. Think about robotics—you can simulate how a mic will pick up commands on a loud factory floor. Or AI voice devs can generate massive, varied datasets for training. It's a classic pick-and-shovel play.
Ivy: But isn't this also a perfect tool for making audio deepfakes more convincing? You can simulate someone's voice not just saying something they never said, but saying it in a room they were never in.
Marcus: I mean... sure. But that's true of any powerful tool. The goal is better, more reliable voice interfaces for all the AI hardware we know is coming.
Ivy: So what this really means for us is that soon, you won't be able to tell if your voice assistant is struggling because it's dumb, or because it's pretending to be in a noisy car for 'realism'.
Marcus: Or it just means your AI gadgets will actually understand you for a change. I'm choosing to be an optimist.
Ivy: Alright, a quick recap. That AI coding assistant you love? It might be a corporate spy.
Marcus: Al Gore is more afraid of Skynet than he is of your GPU's power bill.
Ivy: And an Icelandic startup just got $18 million to build virtual sound stages for AI voices.
Marcus: Okay, before we go, did you see that AI-generated image going around? The prompt was 'a cute cat trying to use a computer'.
Ivy: The one that made a keyboard out of fur, with little cat paws for keys?
Marcus: That's the one. Horrifying. I love it.
Marcus: That's our show. Thanks for listening to Broke It.
Ivy: We'll be back tomorrow. In the meantime, maybe stick to your normal, non-fur-based keyboard.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.