Ivy: An AI just found and used a brand-new, never-before-seen software vulnerability... and it did it in less than an hour.
Marcus: [beat] Yeah. That actually just happened. I'm Marcus.
Ivy: And I'm Ivy.
Marcus: And this is your AI briefing for Sunday, September 6th. We've got a lot to get to, and that was just a taste.
Ivy: So first, we'll get into what that was you just heard—what OpenAI's new GPT-6 model can do, and why security experts are... well, 'concerned' is putting it mildly.
Marcus: Then, the copyright wars heat up again. The Seattle Times and Newsday are the latest newsrooms to sue OpenAI and Microsoft.
Ivy: And later, a cautionary tale from the trails: why you probably shouldn't let Google Gemini plan your next hike.
Marcus: Alright, let's start there. OpenAI just dropped benchmarks for GPT-6 Astra, and the headline feature is... autonomous hacking.
Ivy: [dry] That’s a gentle way to put it. Let's be clear: for the first time, an AI can find and use unknown software bugs, all by itself. That’s a line people have been terrified of crossing for years.
Marcus: But the capability is incredible! This is automated red-teaming. An AI finding holes in your code before the bad guys do. It’s a game-changer for cybersecurity.
Ivy: It's a game-changer for offense, too, Marcus. That's the point. This isn't about using old exploits; it’s about creating new ones. What happens when this gets out? Or when someone else builds one?
Marcus: OpenAI says they're keeping it under lock and key, that they're a 'responsible actor'—
Ivy: A 'responsible actor' that just built a zero-day exploit generator? Forgive me if I'm not comforted.
Marcus: But look at the results! It beat 85% of human security experts in a head-to-head competition.
Ivy: And that's what's terrifying! It's superhuman at the one thing that has always limited cyberattacks. We could go from a few hundred new exploits a year to millions.
Marcus: So what does this mean for a developer or a business right now?
Ivy: It means you can no longer assume your bugs are hidden. Assume they will be found, and fast. You have to shift from reacting to predicting. And right now, nobody knows how to do that.
Marcus: [beat] Unless you have an AI defender that's just as fast. The arms race just went autonomous.
Ivy: An arms race we didn't need to start. Unbelievable.
Marcus: Next up, the legal battles just keep piling on. Now The Seattle Times and Newsday are suing OpenAI and Microsoft.
Ivy: Good. They're joining The New York Times, The Intercept, and a ton of others, all saying their work was stolen to train these models, with no permission and no pay.
Marcus: The AI labs have the same defense: it's fair use. They're not reprinting articles, they're learning from them to make something new.
Ivy: It stops being 'transformative' and starts being 'replacement' when the AI spits out a summary so good you never click the link. That’s the heart of the argument.
Marcus: But isn't that what search engines have been doing for 20 years with snippets?
Ivy: A snippet gives you a taste to make you click. An AI answer digests the whole meal and serves it up as its own. It's not the same thing at all. And they're suing for billions.
Marcus: Billions... [sighs] It feels like trying to put the genie back in the bottle.
Ivy: They're not trying to put the genie back in the bottle. They're trying to make the genie pay rent. News outlets spend millions on reporting. The AI companies used that work to build a trillion-dollar industry and paid zero for it.
Marcus: So what does this mean for us? Does ChatGPT get dumber if they have to pull all that news data?
Ivy: It means the free lunch could be over. Either the models get worse, or the price of licensing all that data gets passed on to you. There's no version of this where the people who made the content get nothing.
Marcus: Unless the courts rule it's fair use. This is headed for the Supreme Court, for sure.
Ivy: And in the meantime, the places that actually produce information are going broke. Great for society.
Ivy: Alright, for our last story... if you're going hiking... maybe stick to a paper map.
Marcus: [laughs] I saw this. Some hikers in California needed rescuing after they let Google's Gemini plan their trip.
Ivy: The sheriff's office put out a notice saying, and I'm quoting, Gemini 'advised them to bring far less food and water than their group required.'
Marcus: No way.
Ivy: They found them miles from the trailhead, dehydrated and exhausted. The AI basically invented a much shorter, easier trail and told them to pack for that.
Marcus: Okay, that's bad, but come on, that’s also user error, isn't it? You don't blindly trust one source—AI or not—for something that important. You check other sources!
Ivy: Do you really? Or do you trust the giant tech company that sells its product as an all-knowing oracle? Google's ads say 'Plan your next vacation with Gemini!'
Marcus: For booking flights! Not for calculating how much water you need in the desert! There’s a disclaimer on these things for a reason.
Ivy: That nobody reads! And that won't help you when you have heatstroke because your pocket genius said one bottle of water was enough. These are the real-world consequences of 'moving fast and breaking things.'
Marcus: So the takeaway here is... pretty obvious, right?
Ivy: The takeaway is that large language models are not reliable. Especially when your health is on the line. They are very convincing liars. Don't trust them with anything that matters.
Marcus: [beat] Or at least, pack extra water.
Marcus: Alright, let's do a quick recap. OpenAI's new model can hack, and everyone is trying to figure out what that means.
Ivy: The Seattle Times and Newsday are the latest to sue OpenAI and Microsoft, demanding to be paid for their work.
Marcus: And finally, maybe don't use Google Gemini to plan a hike, unless you enjoy being rescued.
Ivy: So before we go, did you see that new AI movie critic? The one trained on thousands of reviews?
Marcus: Oh, the one that gave 'Citizen Kane' a 4 out of 10 because it was 'in black and white and the plot was a little slow'?
Ivy: [laughs] That's the one. It perfectly mimics a bad reviewer with zero understanding of art or context. A true masterpiece of mediocrity.
Marcus: And that's our show. We'll be back tomorrow.
Ivy: Until then, stay safe... and please, pack your own snacks.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.