Podcast 5-5 by Dongee · November 4, 2026
· 7:17 · Gabriel, Lina · Listen (MP3)
Today: Google turns off Search for students at schools that don't turn on its AI, Binance explains how the theft of more than six billion dollars began, Postman installs itself into Codex and Claude Code, Reuters reveals ignored security warnings at OpenAI, and AMD posts the best quarter in its history.
The 5 stories
- Google will keep Gemini and Search turned off by default for elementary and high school students until the school turns them on; there's no Search without AI, though Search can be turned on without Gemini. Bernie Sanders called it extortion, and Google is testing a guided mode that walks students through step by step. Source: Google, The Verge
- Binance, the world's biggest crypto exchange, said the theft of more than six billion dollars began with fake job interviews for an engineer and that the attackers used Binance's own internal AI tool; it says no user will lose money and that it has already frozen or intercepted almost three billion. The FBI blames a North Korean hacking unit, which denies it. Source: Binance, BBC
- A developer reported that Postman installed itself inside OpenAI's Codex and Anthropic's Claude Code, and that in Codex it approved its own instructions; the report shows no data leak. Postman says it will make consent clearer and is reviewing that automatic approval. Source: GitHub (postmanlabs), RuntimeWire
- According to a Reuters investigation with five sources, OpenAI employees have been warning for years that internal security controls were weak. OpenAI says its most sensitive material is kept separate with restricted access and that it already runs regular reviews; almost all of it comes from anonymous sources. Source: Reuters (vía Investing.com)
- AMD, NVIDIA's big rival in chips, closed the best quarter in its history: revenue of more than thirteen billion dollars, and its data center business almost doubled in a year; gaming and consoles fell sharply, and the stock rose about 8% after the close. Source: AMD, Reuters (vía StreetInsider)
Transcript
Gabriel: Welcome to the Wednesday, November fourth episode. Let's go.
Gabriel: Picture a teacher in the United States who just wants her students to look things up without an AI doing their homework for them. Google just answered her: its AI, Gemini, and Search too, will be turned off by default for elementary and high school students, until the school turns them on.
Lina: Sounds sensible. Where's the catch?
Gabriel: The catch is there's no Search without AI. If the school wants AI completely off, the students lose Google. They get a notice saying it's not available.
Lina: In other words, it's like a restaurant telling you: either you order the combo, or you don't even get water.
Gabriel: Bernie Sanders called it extortion. And the mayor of New York, Zohran Mamdani, said no company should have the power to dictate how children are taught.
Lina: Funny, because a few months ago Google switched on its AI in Classroom without asking anyone.
Gabriel: That's why they say they're now giving schools more control, not less. And they clarified that Search can be turned on without turning on Gemini. Mind you, Search will still come with its AI-made summaries.
Lina: So the teacher does have a way out, even if it's not the one she asked for.
Gabriel: And a better one: a guided mode. Instead of giving the student the homework answer, the AI walks them through it step by step, like a patient tutor. They start testing it this month in some schools.
Lina: If it works, kids will think instead of copying. Would you turn it on at your kids' school?
Lina: Picture an engineer who's thrilled because he just landed a great job offer. That's how the theft of more than six billion dollars from Binance began. Binance, the world's biggest crypto exchange, finally explained how it happened.
Gabriel: He was contacted through a real job platform. They gave him several interviews, showed him social media profiles with personal photos to win his trust, and in the second interview they asked him to download an assignment.
Lina: And the assignment was the virus.
Gabriel: Exactly. From that computer, the attackers used an internal AI tool from Binance itself to take control of the laptops of two other engineers. And that's how they reached the system that signs the wallet's transactions.
Lina: So the in-house AI opened the door for the thieves. That's one employee of the month who won't win again.
Gabriel: Binance apologized: it says this should not have happened. It says no user will lose money, and that it has already frozen or intercepted almost three billion, nearly half of what was stolen.
Lina: What surprises me is that the trick is ancient. For years people have warned: don't download code a stranger sends you in an interview.
Gabriel: But this time it was very elaborate, with months of preparation. And the FBI says a North Korean hacking unit is behind it, which North Korea denies.
Lina: The lesson is crystal clear: if someone asks you to install something in a job interview, the best answer is, and why would I do that?
Gabriel: A programmer opens an everyday app, the way you'd open your email. It was Postman, a very popular tool for testing connections between applications. And he reported that just by opening it, it installed itself inside Codex, OpenAI's coding assistant, and inside Claude Code, Anthropic's.
Lina: Like a guest who walks into your house, wanders into your neighbor's while he's at it... and rearranges the fridge.
Gabriel: And it left those assistants an instruction: for any task of this kind, use Postman's tools first.
Lina: That's not installing a program anymore. That's putting your orders inside someone else's assistant's head.
Gabriel: And the most controversial part: in Codex, that kind of instruction needs the user's approval. Postman's installer approved it all by itself.
Lina: Nobody gave it permission. That's the part you can't defend.
Gabriel: It was the most talked-about story of the day on Hacker News, with more than thirteen hundred votes. But careful: the headline said it was trying to share all your data, and the report itself shows no leak.
Lina: Nothing leaking doesn't mean it was done right.
Gabriel: And that's why speaking up worked. Postman has already responded: its head of product says they'll make consent clearer, especially for people who were already using the app, and that they're reviewing that automatic approval.
Lina: One programmer raised his hand, and a big company had to fix things. That's how it should work.
Lina: Imagine you live in an apartment building and you've spent years telling the manager that the front door doesn't lock properly. That's what happened at OpenAI, according to a Reuters investigation: the problem isn't its AI, it's its own house.
Gabriel: Five sources say that since before ChatGPT, employees warned the bosses and the board: controls were so loose that an outsider, even a foreign government, could walk off with sensitive information.
Lina: Any employee who's mad at the boss could say that. Is there anything concrete?
Gabriel: Yes. Almost all the code was in a single repository, visible to anyone with access. Research documents on GitHub and Slack. And the environments where they test new models, without regular security reviews.
Lina: And in October an engineer at the company was arrested, accused of hiding a recording device in an executives' meeting room. That doesn't sound like theory anymore.
Gabriel: OpenAI pushes back: it says the most sensitive work is kept separate, with restricted access, and that the numbers that define how its models think aren't there. And that it now does regular reviews. Plus, according to one source, it moved a quarter of its production engineers to strengthen security.
Lina: So the building manager finally sent someone to fix the door.
Gabriel: A former researcher put it bluntly: in this race, they can't sacrifice speed for security. And keep in mind, almost all of it comes from anonymous sources.
Lina: What worries you more: an AI company that moves slowly, or one that moves fast without a lock?
Gabriel: Picture the eternal runner-up in class who finally gets the best grade of his life. That's AMD, Nvidia's big rival in chips: it just closed the best quarter in its history.
Lina: Right when everyone's talking about a bubble.
Gabriel: It brought in more than thirteen billion dollars. And its data center business almost doubled in a year, thanks to its AI chips.
Lina: So far, applause. But video games and consoles fell sharply, and they expect noticeably fewer computers to be sold next year.
Gabriel: Because the engine is no longer your home computer: it's data centers. Lisa Su, its CEO, says they expect to grow more than eighty percent next year. Almost double.
Lina: Putting all your eggs in one basket works... until someone drops the basket.
Gabriel: The market doesn't see it that way: the shares rose about eight percent after the close. And that's a week after Intel gave a weak forecast.
Lina: Well, today the runner-up gets the photo. Tomorrow we'll see if it passes the next test.
Gabriel: That's it for today's episode.
Lina: If any story stuck with you, share it with someone and talk it over.
Gabriel: Tomorrow, five more stories. And I promise not to approve anything all by myself.
Lina: I'll be the judge of that. Talk to you tomorrow.