Ivy: ...silently opted every single developer back into sharing their proprietary code.
Marcus: I'm Marcus.
Ivy: And I'm Ivy.
Marcus: And you're listening to Broke It, your daily AI news for Wednesday, November 4th, 2026.
Ivy: We've got three stories for you today, and they feel... pretty heavy.
Marcus: Starting with OpenAI, who seems to have flipped a very important data sharing switch for all its coding assistant users... without telling them.
Ivy: Then we'll get into a new system from Anthropic that's meant to prove an AI is who it says it is.
Marcus: And we'll end on a tough question: is AI permanently sealing the door on entry-level software jobs?
Marcus: Alright, let's start there. Back to what you said in the cold open. Over the last 24 hours, developers on Hacker News and Reddit started noticing something... terrifying.
Ivy: Terrifying is the right word. It looks like OpenAI silently re-enabled data sharing in its Codex API and tools. That's the setting that lets them use your code—and your clients' code—to train their models.
Marcus: Wait, the one everyone rushed to turn off years ago? That was a huge deal! You had to opt out to protect your IP.
Ivy: And now users are reporting the setting is just... flipped back on. No consent, no notification.
Marcus: Whoa. That is... not okay.
Ivy: It's a colossal IP risk. Imagine you're writing proprietary code for a Fortune 500 company. You opted out. Now, for who knows how long, that code has been feeding straight into OpenAI's next model.
Marcus: Could it just be a bug? Like a database migration gone wrong? I mean, it's happened before.
Ivy: 'It was a bug' isn't a defense, Marcus. Not when you're handling the crown jewels of every company on Earth. At this scale, there's no difference between a bug and gross negligence.
Marcus: [sighs] You're right. The trust is just gone. Again.
Ivy: It's classic 'move fast and break things,' except the 'thing' they broke is the fundamental trust of their customers.
Marcus: So what's the takeaway for developers using these tools?
Ivy: It means stop what you're doing, go to your OpenAI account settings right now, and check. Just assume it's on. And then maybe ask yourself if the convenience is worth this.
Marcus: Oof. A self-inflicted wound for OpenAI.
Ivy: Alright, let's talk about trust from another angle. How do you actually know an AI is the one it claims to be?
Marcus: Right, like how do I know this response is actually from Claude 5 and not some cheaper, knock-off model pretending to be it?
Ivy: Exactly. And Anthropic is working on a system called FlashProve to solve it. TechCrunch reports it's a way to cryptographically verify a model's identity.
Marcus: That's cool. So it's like a watermark for the model itself? A third-party app can't just claim it's using a high-end model and secretly use a cheap one?
Ivy: That's the idea. You get proof of authenticity for any response, so you can be sure the output came from the model it claims.
Marcus: This is huge for fighting misinformation! You could check if a scary government alert actually came from an official model, or if it's a fake.
Ivy: Okay, hold on. It's a noble goal, but who's actually going to check? This sounds like something an enterprise client cares about for audits. Is a normal person ever going to run a cryptographic check on a chatbot response?
Marcus: But tools could build it in automatically! Like the little green lock on a secure website. It builds trust!
Ivy: [dry] We've seen how well people pay attention to browser warnings. The problem isn't usually proving something is fake. It's getting people to care enough to check. And they usually don't.
Marcus: I think you're being too cynical. It's foundational infrastructure!
Ivy: So, soon you'll have a new, technically fascinating way to confirm you're not being lied to about which multi-billion-dollar AI you're talking to. Whether that actually helps is another question.
Marcus: Our last story... this one's tough. A post is blowing up on Hacker News called 'The Ground Floor Is Sealed.'
Ivy: Yeah, I saw that. The argument is AI assistants are now so good at basic coding tasks—writing tests, refactoring, boilerplate—that the whole 'junior developer' role is effectively gone.
Marcus: And it makes a terrifying amount of sense. The work we'd give a fresh grad to get them started... an AI can do it faster and cheaper. A senior dev can just prompt an AI for the first draft.
Ivy: Which erases the entire on-ramp to a software career. If you can't get that first job, how do you ever become a senior developer? We're not just losing jobs, we're losing the entire pipeline for future talent.
Marcus: But isn't this just evolution? We don't have switchboard operators anymore. New jobs will emerge. People will just learn to be 'AI-first' developers from day one.
Ivy: Nice theory, but how do you get good without doing the basics? You can't learn to be an architect without ever laying a brick.
Marcus: So you're saying the ladder is just... gone?
Ivy: I'm saying the first rung is gone. And if you can't get on the ladder, you can't climb. This has huge implications for diversity, social mobility... for a whole generation who was told 'learn to code' was the golden ticket.
Marcus: Wow. So if you're an aspiring developer... what then? Just give up?
Ivy: No. It means the bar is just impossibly high now. You can't just know how to code. You have to be an expert prompter, a systems thinker, and a product manager, all from day one. The ground floor is now the tenth floor.
Marcus: The ground floor is sealed. Yikes.
Marcus: Okay, that was a lot. So, to recap: OpenAI might have just torched user trust by re-enabling data sharing, putting developer IP at risk.
Ivy: Anthropic is building a tool to prove AI identity, but we'll see if anyone uses it. And the traditional entry-level developer job might be going extinct, thanks to AI.
Marcus: Alright, before we go... on a lighter note, did you see that new model trained to heckle stand-up comics?
Ivy: The one that analyzes a comic's pacing to generate 'context-aware' interruptions? [dry] Yes. Finally, an AI that replicates the worst person in the room.
Marcus: [laughs] I can just hear it now. 'Your delivery has 30% less confidence than your opening joke!' Brutal.
Marcus: And that's our show. You've been listening to Broke It.
Ivy: We'll be back tomorrow. Meantime, go check your OpenAI settings. 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.