Ivy: Your boss's algorithm can watch you, judge you... but as of today in California, it can't fire you. At least, not on its own.
Marcus: I'm Marcus!
Ivy: And I'm Ivy.
Marcus: And you're listening to Broke It, your daily AI deep dive. It’s Thursday, October 1st, 2026. Coming up on the show...
Ivy: A new California law puts a human back in the loop for AI-driven firings.
Marcus: A cautionary tale: AI coding assistants are leaking corporate secrets all over GitHub.
Ivy: And OpenAI just dropped GPT-6. Are their new 'Sol' and 'Luna' models finally AI for the rest of us?
Marcus: Let's start in California, Ivy. This is a huge win for worker protections.
Ivy: It's a first-in-the-nation law. Governor Newsom signed a bill that says an 'automated-decision tool' can't be the only reason you get fired.
Marcus: 'Sole factor'—that's the key. So an AI can still flag you for 'low productivity,' but an actual human has to sign off on the pink slip.
Ivy: Precisely. The law mandates 'meaningful human oversight.' And 'meaningful' is a word lawyers are going to have a lot of fun with.
Marcus: But it's a start! No more Amazon workers getting fired by an app with zero recourse. It’s a baseline for human dignity.
Ivy: A floor that's still in the basement, perhaps. The law also covers algorithmic surveillance. Employers now have to notify workers about the data they're collecting and what it's used for. But they can still... you know, do it.
Marcus: Transparency is the first step, though! You can't fight what you can't see.
Ivy: The flip side is this just becomes a rubber-stamp. A manager gets a list of fifty names from the algorithm, they're busy, they click 'approve.' How 'meaningful' is that, really?
Marcus: That's too cynical. It creates legal accountability. If a manager just rubber-stamps a biased firing, the company is on the hook now. It forces them to actually think.
Ivy: We'll see. The proof will be in the first wrongful termination lawsuit that cites this law. Until then, it's a strongly worded press release.
Marcus: So for our listeners, if you're in California, your job just got a little safer from robo-bosses. And for everyone else? This sets a precedent that other states are likely to follow.
Ivy: Or, expect companies to find creative new ways to define 'meaningful human oversight.' Like having an intern check a box.
Ivy: Next up: what happens when your helpful AI assistant has loose lips? Or in this case, leaky code.
Marcus: This is a developer's nightmare. We're talking about AI coding agents, the tools that help you write and debug...
Ivy: ...are getting a little too helpful. A new report found they're pushing code to public GitHub repositories by default.
Marcus: By default? That's insane. So a developer is working on a new internal feature, using an AI to help, and that AI just... posts the work-in-progress to the entire world?
Ivy: Exactly. The report found over 13,000 instances of sensitive data being exposed. We're talking API keys, database credentials, chunks of proprietary algorithms...
Marcus: Wait, actual hardcoded secrets? Not just boilerplate code?
Ivy: Oh yeah. We're talking private keys for AWS, Google Cloud... a goldmine for attackers. And it's happening because developers are just hitting 'run' without checking the settings.
Marcus: It's the classic 'move fast and break things' problem. Everyone's so excited about the new toy they forget basic security.
Ivy: It's more than that. It's a failure of the tool's design. The default for anything that touches proprietary code should be 'private.' Opt-in to public, not opt-out.
Marcus: Totally. It's a rookie mistake by the toolmakers. The takeaway for our listeners who code: check your settings. Now. And treat these AI tools like an untrusted intern until they earn your trust.
Ivy: And for managers, it means you need a policy for using these agents yesterday. Don't just let your team run wild with them.
Marcus: This is the kind of mistake that doesn't just cost you money, it could sink a company. Imagine your secret sauce getting pasted onto a public forum...
Ivy: ...where it's immediately scraped, indexed, and probably incorporated into the next generation of AI models.
Marcus: And finally, OpenAI is back. They just announced their GPT-6 family of models—called Sol and Luna—and I'm pretty excited about this.
Ivy: Another major release. It feels like we just got GPT-5. So what's the hook this time?
Marcus: The hook is splitting it into tiers. You have Sol—that's the big, powerful, top-of-the-line model. It's supposedly their most capable ever, but it's going to be expensive.
Ivy: The 'pro' model that only enterprises can afford. Got it. So what's Luna?
Marcus: Luna is where it gets interesting. It's a family of smaller, efficient models. OpenAI claims the best Luna model is as good as GPT-5, but at a fraction of the cost. And they scale all the way down to models that can run on your phone.
Ivy: On-device, you say? So, no more sending all my data to OpenAI's servers for every little query?
Marcus: That's the promise. A truly private, fast, and smart AI assistant on your phone. For developers, it means cheaper, faster API calls for most jobs. You only call in the big 'Sol' model when you really need it.
Ivy: Okay, I'll admit, a tiered system makes sense. The 'one model to rule them all' approach was always inefficient. But the proof is in the pudding. Their blog post is full of impressive-looking charts, but they're OpenAI's charts. We need independent benchmarks.
Marcus: Totally. But if those claims hold up, it's a game-changer. It makes powerful AI accessible to everyone from small startups to hobbyists. It's democratizing.
Ivy: Or, it creates a new class system for AI. The rich get the 'Sol' intelligence, the truly groundbreaking stuff, and the rest of us get 'Luna,' the good-enough version that's a generation behind.
Marcus: I don't see it that way—it's about using the right tool for the job! You don't need a sledgehammer to hang a picture.
Ivy: Unless the picture is a 500-pound oil painting. My point is, the gap between the top model and the rest will be telling.
Marcus: Okay, fair. So the takeaway: Developers, get ready for cheaper API bills. Users, get ready for smarter, faster AI in your apps. And maybe, just maybe, an AI that lives right on your phone.
Ivy: What it really means is that we'll be spending the next month sifting through benchmarks to see if Luna is actually as good as last year's flagship, or if it's just clever marketing.
Marcus: Alright, let's recap the headlines on Broke It.
Ivy: California passed a law requiring a human to be involved when an AI tries to fire you.
Marcus: AI coding tools are leaking company secrets to GitHub. Go check your settings.
Ivy: And OpenAI launched its new GPT-6 Sol and Luna models, promising more affordable and practical AI for everyone. We'll see.
Marcus: Before we go, one more thought on GPT-6... Sol and Luna. Sun and Moon. I love the names.
Ivy: I find it a bit on the nose. The big, powerful, expensive one is the Sun. The smaller, cheaper, 'for the people' one is the Moon, which just reflects the Sun's light.
Marcus: Oh, wow. I didn't even think of that. You just can't help but find the cynical angle, can you?
Ivy: It's not cynicism, it's literary analysis. They're telling you the power dynamic right there in the branding.
Marcus: And on that note of cosmic cynicism, that's our show for Thursday, October 1st!
Ivy: Thanks for listening to Broke It. We're back tomorrow with more from the world of AI.
Marcus: Yeah, don't let the robo-bosses, leaky code, or cosmic cynicism get you down. Talk to you tomorrow!
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.