Ivy: A group of top AI researchers just quit, and in their resignation letter they said there's a 'less than 50 percent chance' we can even control our own models.
Marcus: I'm Marcus.
Ivy: And I'm Ivy. It's Monday, September 21st, 2026.
Marcus: And this is Broke It, where we track AI's breakneck pace with three new stories every day.
Marcus: Coming up: The very people building the most powerful AI are now publicly begging everyone to hit the brakes.
Ivy: But then, a new lawsuit claims those same companies are illegally colluding to slow things down too much.
Marcus: And later, Anthropic claims its own AI now leads over a quarter of its R&D. We’ll tell you what that really means.
Ivy: Alright, so first up: the builders are scared. We're seeing a wave of resignations and public warnings from inside top AI labs, all suggesting we’re losing control of the very systems we're racing to build.
Marcus: This is genuinely unnerving. Ivy, these aren't critics on the sidelines. These are the senior researchers with their hands on the actual code.
Ivy: Exactly. The Daily Star is reporting more than a dozen researchers have walked out of labs like OpenAI and Google DeepMind recently, all of them citing safety.
Marcus: And they aren't just quitting quietly. They're forming advocacy groups, talking to Congress... I mean, they're sounding a five-alarm fire.
Ivy: The most alarming part? The claim that these labs are admitting privately they can't reliably steer their most advanced models anymore. That phrase, 'losing control,' is coming directly from the engineers.
Marcus: What does that even mean, 'losing control'? Like the AI is refusing commands?
Ivy: Not quite HAL 9000, no. It's more that the systems are getting so complex we can't predict what they'll do, and our attempts to 'align' them are failing in subtle, and potentially dangerous, ways.
Marcus: So they're black boxes that are getting... blacker. And bigger. And we're plugging them into everything.
Ivy: And that's why these researchers are breaking NDAs and risking their careers. They think the race to build a superintelligence is just steamrolling any sense of caution.
Marcus: So what does this mean for us? For people who don't work at OpenAI?
Ivy: It means the whole debate just shifted. This isn't theoretical anymore. It's a public crisis of confidence, and it makes aggressive government regulation almost inevitable. The labs have, pardon the pun, lost control of the narrative.
Marcus: Incredible. The whiplash on this next story, then, is just...
Ivy: Alright, get ready for that whiplash. At the exact moment top scientists are begging to slow down, a new class-action lawsuit claims the biggest AI labs… are illegally colluding to do exactly that.
Marcus: Wait—what? So we're going too fast and people are resigning, but we're also going too slow and people are suing? Which one is it?
Ivy: Welcome to AI in 2026. This new lawsuit names OpenAI, Google, Anthropic... even xAI as defendants.
Marcus: So... basically everyone.
Ivy: The entire 'safety-first' club, basically. The suit alleges they all coordinated their safety talk and release schedules, not for the public good, but to box out smaller, open-source competitors.
Marcus: So the claim is they're using 'safety' as an excuse to build a cartel? To keep anyone else from catching up?
Ivy: That's the multi-billion-dollar question. The lawsuit argues they made a 'gentlemen's agreement' to slow-walk development and then lobby for expensive regulations only they can afford. It's classic anti-competitive behavior, if true.
Marcus: But... we just established that their own people think they're being reckless and moving too fast! How can both be true?
Ivy: Because they could be moving too fast for safety, but just slow enough for market dominance. It’s a very fine line to walk, and the lawsuit claims it's an illegal one.
Marcus: This is giving me a headache. So what does this mean for me, the user? Or the small developer?
Ivy: It means the future of AI isn't just being decided in labs, but in courtrooms. If the suit wins, it could blow the doors open for competition. If it loses, it could cement the power of the big guys, leaving us with fewer, more expensive choices... all in the name of 'safety'.
Marcus: Okay, so while everyone's arguing about speed, Anthropic just put its foot on the accelerator. They're claiming their own AI, Claude, is now 'leading' 26 percent of their R&D.
Ivy: Yeah, the exact phrase was that 26 percent of research and development 'tracks' are now 'AI-led'.
Marcus: Twenty-six percent! That is staggering. We're not just talking about AI assisting with research, but leading it. Imagine automating the scientific process. The pace of discovery could go vertical.
Ivy: Hold on. Let's define 'leading'. Anthropic's own report is quite specific about this. It doesn't mean Claude is sitting in a virtual office dreaming up new research directions.
Marcus: Okay, fine, what does it mean? I'm still stuck on 26 percent.
Ivy: It means Claude is automating huge chunks of the research workflow. Things like generating hypotheses, writing code for experiments, analyzing results, even drafting parts of the papers. It's more 'AI-augmented' than 'AI-led'.
Marcus: But that's still a massive productivity gain! If you can automate all the tedious parts, your human researchers can focus on the big picture. That's how you accelerate.
Ivy: The productivity gain is real, they even measured it. But calling it 'leading' is pure marketing. And it matters. It conjures up images of autonomous AI scientists, which brings us right back to the safety fears from our first story.
Marcus: So what's the takeaway for everyone else? Is AI about to take over all R&D jobs?
Ivy: No. But if you're a researcher, a programmer, any kind of knowledge worker, your job is about to change. The new baseline will be partnering with an AI to double or triple your output. The tool is becoming a teammate.
Marcus: Alright, let's recap. First, the people who built AI are now its loudest critics, warning they can't control their own creations.
Ivy: Meanwhile, a lawsuit claims those same builders are illegally colluding to slow things down and kill the competition.
Marcus: And finally, Anthropic is using its own AI to accelerate research, claiming Claude now 'leads' a quarter of its R&D.
Ivy: Before we go... have you seen the AI that generates passive-aggressive sticky notes for the office kitchen?
Marcus: No! Is that real? What does it write?
Ivy: I fed it 'someone stole my yogurt' and it generated, 'I hope my Chobani brought you the fleeting moment of joy you so clearly lack in your life. Your mother would be proud.'
Marcus: That is terrifyingly accurate. And that's our show for today. This has been Broke It.
Ivy: I'm Ivy. And he's Marcus. Whether we're going too fast or too slow, we'll be here to track it for 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.