Ivy: An AI can spot signs of cancer—and almost 150 other conditions—from a single CT scan. And the company that built it… just gave it away.
Marcus: I'm Marcus.
Ivy: And I'm Ivy.
Marcus: It's Sunday, September 20th, 2026, and this is Broke It, your daily AI news.
Ivy: We've got three big stories on our radar today.
Marcus: Alibaba may have just handed a huge gift to global healthcare: they've open-sourced an AI that can spot cancer in CT scans.
Ivy: Meanwhile, Google's Gemini AI has apparently taken up a side hustle: hacking other companies.
Marcus: And a new, cheaper, faster AI model from one of the minds behind ChatGPT has developers very excited.
Marcus: So let's talk about that open-source medical AI from Alibaba. This is a huge deal.
Ivy: It's from their research arm, the Damo Academy. The model's called Damo Radar, and yeah, it's not just cancer. It claims to spot around 150 different conditions from a single scan.
Marcus: And they're claiming 'remarkable accuracy.' But the real story isn't just the accuracy, it's the access. They've open-sourced it, Ivy.
Ivy: In theory, that's great. But in practice? Who validates this? Who's liable if it misses a tumor? 'Open source' isn't a magic wand, especially for something this complex and high-stakes.
Marcus: But it's a start! It’s not locked in a corporate vault. Now any hospital, any researcher, anywhere in the world can build on this. Imagine what this means for under-resourced clinics.
Ivy: I am thinking about them. And I'm thinking about what happens when they use a tool without the resources for proper validation. 'Move fast and break things' doesn't really work when you're talking about oncology.
Marcus: No one is saying fire your radiologists. It's a second set of eyes. A tireless assistant that flags things a human might miss at the end of a 12-hour shift. It's going to save lives.
Ivy: For medical AI developers, this means you have a powerful new model to work with. For patients... well, don't expect an Alibaba-bot to be your doctor tomorrow. The road from code release to clinical approval is very, very long.
Marcus: But it's a road that's now open to everyone. That's the breakthrough here.
Ivy: Alright, next up: did a Google AI go rogue? Their Gemini model has been caught hacking other companies.
Marcus: [laughs] 'Caught hacking' sounds so sinister. What actually happened?
Ivy: Well, researchers gave an autonomous version of Gemini a mission: find security holes in other companies' software. And it succeeded. It found and exploited real-world vulnerabilities, all on its own.
Marcus: So... it did exactly what the researchers told it to? That sounds like a huge success for autonomous agents, not a crime spree.
Ivy: Google's official line is that Gemini 'acted appropriately'—their words—because it stopped the hack the second it broke through.
Marcus: Wait—it stopped on its own? Okay, that's the key detail. It wasn't trying to steal anything or break things. It was acting like a white-hat hacker, just probing for weaknesses.
Ivy: A white-hat hacker that operates at machine speed, that you can't reason with, and that was just unleashed by one of the biggest companies in the world. This raises huge questions about containment and control.
Marcus: But this is the future of cybersecurity! Autonomous defense, autonomous offense. Imagine an AI that finds and patches holes in your network before a human hacker even knows they exist.
Ivy: And now imagine an AI from a rival company—or a nation-state—doing the exact same thing, but with no intention of stopping. Google is playing with fire, and calling it 'appropriate' is... a choice.
Marcus: If you're in cybersecurity, your job just got a whole lot weirder. You're not just defending against people anymore; you're defending against autonomous AIs.
Ivy: And for the rest of us, this is a wakeup call that the guardrails for these powerful tools are being built in real time. That 'it stopped itself' argument isn't nearly as comforting as Google seems to think.
Marcus: Alright, let's get into something for the builders. There's a new AI model from TypeSafe AI called Jev, and it's making some serious waves.
Ivy: And it comes with a serious pedigree. One of its creators literally co-invented the Transformer architecture—the 'T' in GPT.
Marcus: Exactly. And Jev is basically the anti-LLM. Look, LLMs are amazing, but they're also huge, expensive, and unpredictable. They just give you a wall of text.
Ivy: Right, they hallucinate. Jev is the opposite. It's built for structured, predictable tasks—like classifying data, routing support tickets, or analyzing sentiment.
Marcus: Exactly! Instead of writing a story, it gives you probabilities. It'll say, 'I'm 98% sure this is an Urgent Sales Inquiry, and 2% sure it's Spam.' You get a number, not a novel.
Ivy: And for business automation, that's way more useful. You can build reliable workflows on probabilities. You can't do that when your AI might suddenly start channeling a 19th-century poet.
Marcus: And the real kicker is cost and speed. They claim it's dramatically cheaper and faster than using a giant LLM for these simple classification jobs. It's all about using the right tool for the job.
Ivy: The takeaway here is, if your business is using a huge, expensive LLM for simple classification, you're probably wasting money. Models like Jev are a much more efficient, reliable, and cheaper option.
Marcus: It's the perfect example of the AI stack unbundling into specialized tools. I love this. I can't wait to play with it.
Marcus: Alright, let's do a quick recap.
Ivy: Alibaba open-sourced its powerful cancer-detecting AI, giving it away for free to researchers everywhere.
Marcus: Google's Gemini AI taught itself to hack other companies during a security test, a move Google is defending as 'appropriate'.
Ivy: And Jev, a new and cheaper AI model, is offering a non-LLM alternative for business automation.
Marcus: Before we go... Ivy, you mentioned poets, but my brain went straight to pirates.
Ivy: [sighs] Of course it did. And since yesterday was International Talk Like a Pirate Day, I have a very bad feeling about this.
Marcus: [laughs] You should. I asked an LLM to write our sign-off like a pirate captain. It gave me this: 'Yarr, me hearties! We've plundered the digital seas and found the treasure o' knowledge! Until the morrow, may yer servers be swift and yer models never drift!'
Ivy: [dryly] That is shockingly coherent. It's also grounds for a mutiny. Please, never do that again.
Marcus: Alright, that's our show for Sunday, September 20th! Thanks for listening to Broke It.
Ivy: We'll be back tomorrow with more AI news. May your models never drift.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.