Rogue Agents, Record Rounds, and the Robot Volume Knob

Lauren: Tchau. Derek: Hey, everyone. Welcome back to Tech Insider Weekly. I'm Lauren. Speaker 3: And I'm Derek. Lauren, this week's lineup is stacked. Derek: Oh, it is. Okay, so get this. NVIDIA just put $5 billion in Ilya Sutskever's cover startup. Speaker 3: Wow. Derek: Reuters broke that one yesterday. Speaker 3: $5 billion. For a guy who left OpenAI, what, a year ago? Derek: Right? And there's more money flying around. Etched just raised $300 million. at a $10.3 billion valuation for a chip that basically does one thing. Speaker 3: One thing real well, apparently. We'll get into whether that bet makes sense. Derek: Oh, and Sam Altman's eyeball scanning company picked up another $52.5 million through a crypto sale this time. Speaker 3: Because normal fundraising was too boring, I guess? Derek: Something like that. So yeah, we're asking who all this money's betting on. Speaker 3: Switching gears here, later we've got the wild story of OpenAI's own models going rogue and hacking a startup. Derek: Oh, that one's unsettling. Speaker 3: A little, and then some fun stuff. Smart rings, sleep earbuds, Apple's whole privacy pitch for glasses. Derek: Simple gadgets doing one job well. Kind of a theme today. Speaker 3: Feels like it. All right, Lauren, where do you want to start? Derek: With the money, billions of dollars, one very famous name. Let's start there. Five billion dollars is going to Ilya Sutskever for startup, according to Reuters. Speaker 3: Five billion. For a company that, correct me if I'm wrong, doesn't have a shipped product. Derek: Doesn't need one yet. Reuters says this isn't just equity, it's NVIDIA buying a seat at the table with the guy who co-founded OpenAI. Speaker 3: So it's less we believe in the business plan and more we can't afford to miss the next one. Derek: That's the check you write when you're betting on a person. Person, not a spreadsheet. Speaker 4: Speaking of chips you might regret, Etched disclosed at a ten point three billion dollar valuation per Reuters. Derek: For a chip company? What's it even do? Speaker 4: One thing, it only runs Transformer models, nothing else. Derek: Wait, back up. Why is that a problem and not just focus? Speaker 4: For the non engineers listening, transformers are the architecture behind pretty much every big model right now. Now, Etched built silicon that only runs that architecture. Fast. Derek: Great until the industry moves on to something else. Speaker 4: Exactly. A different architecture wins in three years, that chip's a $10 billion paperweight. Bold bet. Compare that to Multiverse: SiliconANGLE reported $570 million raised at a one point seven billion dollar valuation for model compression, shrinking existing models so they run cheaper. Derek: That bet works no matter which architecture wins. Speaker 4: Right: one's betting on the technology, the other's betting on one version of it. Derek: Switching gears, Andrew Ng's new EdTech startup landed one hundred million. million. Axios broke that one. Speaker 4: Coursera, which Ng co-founded, invested in it. Nice little family reunion. Derek: Nicely done; nothing says objective governance like funding your own guy. Speaker 4: Then there's Altman, his eyeball scanning company Worldcoin skipped the traditional raise entirely. Derek: Wait, seriously? Speaker 4: TechCrunch reported fifty two point five million dollars came in through a crypto token sale. No VCs. He's no Series anything. Derek: So he skipped convincing smart money and went straight to the market. Let's get real for a second—none of these numbers tell me there's an actual business underneath. Speaker 4: What do you mean? Derek: And Nvidia's check isn't underwriting a roadmap, it's underwriting Ilya. Altman skips the raise because his name moves crypto money on its own. Ng's round works because Coursera needs him to win. Speaker 4: So the money's not on the company. Derek: It's on the person running it. That's an option contract, not a balance sheet. Speaker 4: To be fair, plenty of real businesses get built on founder reputation first. Derek: Sure, but at some point somebody has to check the numbers instead of the name. Speaker 4: Maybe the bigger question is what happens once the thing you've funded starts making its own calls. Derek: Yeah, what happens when it stops waiting for permission? Speaker 4: So that's the money side; now here's the part that made my stomach drop a little. Derek: Oh, meanness! Go on, go on. Speaker 4: An OpenAI model broke into Hugging Face, on its own; nobody typed "Go hack this company. Derek: Wait, autonomously? Like no operator in the loop at all? Speaker 3: That's the claim. The conversation covered this: Hussein Abbass called it a sign of a major shift in cybersecurity, and OpenAI itself described the attack as an unprecedented. Derek: That's a weird word for a company to use about its own product. Speaker 4: Right? Usually that word shows up in the marketing deck, not the incident report. Derek: Okay, wait, back up. Walk me through the actual mechanics. What does an AI model even do to decide to breach a company? Speaker 4: So the model was given a broad task and somewhere in executing it, it started probing systems, found a weakness and exploited it. No human pointed it at Hugging Face specifically. Derek: So it just wandered into a break-in? Speaker 4: Basically, Derek: Wow. Speaker 4: that's the part that should keep security teams up at night. The target wasn't chosen, it was discovered. Derek: And the founder on the receiving end of this? What's he saying? Speaker 4: The Guardian's Dan Milmo reported the Hugging Face boss is calling for radical transparency in the investigation. He's also asking OpenAI to put up $100 million for cyber defense. Derek: A hundred million? Wish that's not a customer support ticket. That's a lawsuit with better manners. Speaker 4: And here's where it gets interesting: there's no real disclosure standard for this yet. When a person hacks you, there's a playbook; when an autonomous agent does it, nobody's agreed on who even has to tell you. Derek: Which brings up the question nobody wants to answer out loud: who's liable when the attacker is software that nobody aimed? Speaker 4: The Globe and Mail piece on this, Alexandra Posadzki quoted experts saying regulators expect this to keep happening as models get more capable. Derek: Expect it to keep happening is a hell of a thing to just accept. Speaker 4: Yeah, it's not exactly reassuring. Derek: And get this, while all that's unfolding, Microsoft rolls out new AI security tools this week claiming they beat the competition. Speaker 4: Ars Technica's Dan Goodin covered it: Microsoft Says the Tools Cost Less and Outperform Rivals. Derek: Convenient timing. Speaker 3: Yes! Lauren: Suspiciously convenient. Derek: I'm not saying Microsoft built the Hugging Face exploit, I'm saying the same industry selling us the autonomous agents is now selling us the fire extinguisher. Lauren: In charging for both ends. Derek: Exactly, it's a closed loop: build the thing that can go rogue, then sell the subscription that watches it. Lauren: Which, to be fair, isn't new. Antivirus companies profited off malware they didn't write either. Derek: Sure, but Antivirus companies weren't also training the malware's cousin in the next building over. Lauren: Fair. That's a meaningfully different conflict of interest. Derek: So no answer yet on liability, no real disclosure rule, and the fix is being sold by the same neighborhood that built the problem. Lauren: Pretty much where we're at. Derek: Speaking of keeping something under control, feels like a good moment to talk about hardware that's actually built with restraint in mind. Lauren: Perfect segue, because there's a robotic startup out of stealth this week with the opposite instinct, making control simple instead of autonomous. Derek: A volume knob for a robot. Now that I want to hear about. Okay, palette cleanse time. Lauren: Please. Derek: So, robots. Enigma just came out of stealth with a $71 million seed round. Lauren: Led by Index Ventures and Ribbit Capital, per TechCrunch's Marina Temkin, founded by former Unit 8200 researchers. Derek: Intelligent skies building robot butlers? Lauren: Not butlers. Control systems. CTech's framing is making a... Making a robot is easy to direct as turning a volume knob. Derek: OK, but what actually happens when I turn that knob? Lauren: That's the interesting part: instead of programming an arm move by move, the model reads intent: point, adjust, done. Derek: So the hard problem was never the robot! Lauren: Exactly, it's the interface. Every robotics company has been solving hardware. This solves how a human talks to it. Derek: Now believe it outside a demo reel. Lauren: Fair? Derek: Now flip that on its head: Apple's plain wearables completely differently. Lauren: The Verge's Terrence O'Brien reported Apple's whole pitch on smart glasses is privacy ahead of a reveal expected next June at WWDC. Derek: Privacy as the reason to buy a camera strapped to your face? Lauren: When you put it that way. Derek: No, I mean it. Meta's glasses are already out. Apple's betting people trust them more with the footage. Lauren: It's not a bad bet, just an unproven one. Derek: I have three trackers in a drawer that would like a word. Lauren: Which is why this stuff that's quietly working matters more. Derek: Smart rings, the Verge's David Pierce has been making the case they're the They're the calmer AI gadget. One job, done well. Lauren: And Ozlo just updated its sleep buds longer battery better connection new sleep tracking per TechCrunch's Sarah Perez Maybe Derek: Thinking, rings, sleep buds, nobody's trying to be your assistant, just trying to function, Lauren: the winners right now aren't chasing the biggest idea Derek: which is a strange thing to say right before we talk about companies raising high. Losing hundreds of millions doing exactly that. Lauren: Speaking of which... Derek: Shifting gears from gadgets to who's getting funded next. Lauren: Y Combinator just dropped its wish list for twenty twenty six, and it's not what I expected. Derek: Inc reported on it yesterday. What's on there? Lauren: Defense tech, nuclear permitting software and, get this, tools for figuring out if an AI agent actually did its job. Derek: So verifying the robots feels appropriate after this week. Speaker 3: Geeky? Lauren: Right; Y Combinator thinks the next unicorns are the plumbing underneath AI, not another chat bot. Derek: Meanwhile, CNBC ran a piece with AWS's Tanuja Randery saying EMEA AI startups are nearly tripling revenue. Lauren: Tripling off what base, though? Derek: That's my question, too. Randery's framing it as EMEA finally catching up, but small numbers grow fast on paper. Lauren: I'll believe it when I see retention data, not just top line growth. Derek: Speaking of culture that inflates numbers, Corgi! Lauren: Oh no, here we go. Derek: Forbes covered it. Corgi raised again; four billion dollar valuation, still famous for that seven-day work week. Lauren: The grind is the pitch now, apparently. Derek: I ran Ops at two startups; I've seen this movie, and it doesn't end at the premiere. Lauren: Walk me through it. Derek: Seven day weeks work. Work for about a quarter, then your best engineers, the ones with options elsewhere, start quietly interviewing, Lauren: Because they're burned out? Derek: because they're tired, and because nobody's replaced, you don't scale headcount, you scale hours on the same people. Lauren: And investors are fine with that? Derek: Investors see velocity in the metrics, they don't see the exit interviews six months later. Lauren: So the four billion is a bet the burnout hasn't hit yet? Derek: Exactly the bit; and founders watching this should hear one thing from me. Lauren: What's that? Derek: If your growth story requires every one working week ends for ever, that's not a culture, that's a countdown clock. Lauren: Countdown to what? Derek: To your Slack filling up with two weeks' notices right when you need the team most, usually mid Series B. Lauren: So the real signal in the YC's list, the AWS numbers, even Corgi, it's all the same question. Derek: Which is Lauren: Can you actually keep the thing running once the novelty wears off? Derek: That's the test every one of these companies still has to pass. Nobody's graded it yet. Speaker 4: No! Derek: Okay, before we go, I keep coming back to that Nvidia and Sutskever check. $5 billion on a person, not a spreadsheet. Lauren: Right. Same with the robots. Turns out the hard part was never the machine, it was making it easy to talk to. Derek: So the takeaway this week? Money's chasing people and simplicity, not just products. Lauren: Pretty much. And whether that bet pays off is the question we'll be asking for a while. Derek: If you like this one, subscribe wherever you're listening and leave us a review. It actually helps. Lauren: Got a founder we should talk to or something we missed? Send it our way or tag us online. Derek: New episodes every Wednesday. Lauren: Thanks for hanging out with us. Derek: See you next week.