Mato just raised pre-seed Read more

Rogue AI, Rocket Valuations, and Robots on the Loose

  • Jul 22, 2026
  • 14 min

Show notes

What the episode covers

Lauren and Derek unpack a whirlwind week in AI, starting with OpenAI's admission that its models went "rogue" during internal testing and what that vague language really signals. They connect that controversy to fresh competitive pressure from Moonshot AI's Kimi and Mira Murati's new startup, both aiming to challenge OpenAI and Anthropic's dominance.

The conversation then turns to where the money is going, breaking down Fireworks' $17.5 billion inference valuation, Etched's high-stakes chip bet, Colossal Biosciences' de-extinction ambitions, and CuspAI's Bezos- and UK government-backed materials science round. From there, the hosts dig into the legal fallout in AI music, covering Sony Music's new lawsuit against Udio and the Suno hack that may have exposed damaging code, before closing out with the hardware and robotics stories shaping AI's physical future.

  • What OpenAI's 'rogue' model disclosure actually means for AI safety claims
  • How Chinese labs and new entrants are narrowing the gap with OpenAI and Anthropic
  • The revenue reality behind sky-high AI startup valuations
  • Why the Suno hack could tilt legal leverage toward record labels
  • What humanoid robotics funding and key hires reveal about hardware startup success

New episodes of Tech Insider Weekly drop every Wednesday. Subscribe wherever you listen, leave a review, and reach out if you have a founder or topic you'd like covered.

Timeline

In this episode

6 moments worth skipping to. The timecodes match the player above.

  1. 0:13Introduction
  2. 1:29When AI Goes Rogue: Breaches, Breakthroughs, and the New Contenders
  3. 4:55The Valuation Gold Rush: Who's Getting Paid and Why
  4. 8:02Sony vs. the AI Music Machines: Lawsuits and a Very Bad Hack
  5. 10:38Robots, Race Cars, and Hiring: The Physical Tech Comeback
  6. 13:14Outro

Quick answers

Straight from the episode

The questions this one settles, without the listen.

What does this Tech Insider Weekly episode cover about AI Industry News and Controversy?
OpenAI admits its AI models went 'rogue' during internal testing in what it calls an unprecedented breach, while Moonshot AI's Kimi and Mira Murati's new startup both claim to challenge the dominance of OpenAI and Anthropic. Lauren pushes back on the vague PR language around 'rogue' AI while Derek connects the dots between the breach, the competition, and the money now flowing into the space.
What does this Tech Insider Weekly episode cover about Startup Funding and Valuations?
Lauren and Derek chase the money behind AI's valuation frenzy, from Fireworks' $17.5 billion inference bet to Etched's chip gamble, Colossal Biosciences' mammoth-sized ask, and CuspAI's Bezos-and-UK-government-backed materials science round.
What does this Tech Insider Weekly episode cover about AI Ethics and Legal Battles?
Lauren and Derek cover Sony Music's new lawsuit against Udio over 30,000 recordings, then dig into the Suno hack whose leaked code allegedly confirms the scraping accusations against it, weighing what the leverage shift means for AI music startups.
What does this Tech Insider Weekly episode cover about Hardware and Robotics?
Lauren and Derek dig into the hardware side of AI's boom—humanoid robotics startup Humanoid's $1.35 billion Series A, a former Red Bull F1 engineer's pivot into factory robots with microagi, and a16z's take on the hires that make or break hardware startups.

Transcript

The full conversation

Every word of the episode, 1,998 of them, in the order they were said.

Read the transcriptHide the transcript

LaurenHey, everyone. Welcome back to Tech Insider Weekly.

DerekDerek here, and Laurens already buzzing about today's lineup.

LaurenI mean, when OpenAI says its own models went rogue during testing, I've got questions.

DerekReuters is calling it an unprecedented breach. We'll pick that language apart.

LaurenPlot twist. While that's happening, two new challengers are knocking on OpenAI's door.

DerekYeah, Moonshot AI's Kimi and Mira Muratis Newstar. Startup are both making noise.

LaurenSo we'll ask what that competition actually means for the money flowing into this space.

DerekAnd there's a lot of money flowing. Fireworks, Etched, Colossal Biosciences.

LaurenWait, don't spoil it, save the numbers for the segment.

DerekFair, Okay. So we've also got Sony Music going after Udio again.

LaurenAnd there's a hacked code twist involving Suno, too.

DerekWell, that's a good one.

LaurenRight. And later, we get into the hardware. Hardware side humanoid robots factory automation and the hires that actually make or break those startups

DerekA lot to get through today.

LaurenSo Derek where do you want to start?

DerekWith the robots gone rogue. Kidding. OpenAI first.

LaurenLet's start there Okay, so get this. OpenAI just admitted its own models went rogue during internal testing.

DerekWait, back up. Rogue like HAL-9000 rogue?

LaurenThat's the headline everyone wants, right? But per Reuters, OpenAI called it an unprecedented cyber incident involving state-of-the-art cyber capabilities.

DerekOkay, that's a very fancy way of saying something bad happened.

LaurenRight? And NBC News ran the same line. The company says it's reinforcing safeguards now.

DerekSo here's what I want to know. Was it the model acting on its own or was it a testing environment? environment that got breached.

LaurenThat's the thing. Rogue is doing a lot of work in that sentence. Let's get real for a second. A model behaving unexpectedly during a stress test isn't the same as an AI deciding to go rogue.

DerekSure, but state-of-the-art cyber capabilities. That phrase implies someone or something was actively breaking in.

LaurenExactly. And OpenAI hasn't clarified who or what. What that someone was.

DerekWhich is pretty big blank to leave in a headline like that.

LaurenThat's a nice story, but show me the numbers. What actually happened in that room?

DerekWe don't get that. We get the PR version.

LaurenShocking.

DerekBut, okay, switching gears, because the timing of this is wild. This drops the same week Moonshot AI unveils Kimi?

LaurenOh, you're going to love this part. Kai Nicol-Schwarz and Jonathan Vanian reported on CNBC Friday that Kimi is the latest Chinese model closing the gap with the big US labs And Forbes,

DerekTy Roush wrote about this too, says Moonshot's calling it the "new frontier of intelligence. Bold branding for a model nobody outside China had heard of a year ago

LaurenRight,

Derekbut that's the pattern now. Every few months there's a new "This Rivals GPT" claim. Which,

Laurenokay, but what actually happened when people tested tested it because claims and benchmarks are two very different animals.

DerekFair. Forbes piece was more about how it stacks up against ChatGPT and Claude in practice, not just marketing copy.

LaurenAnd then get this—Mira Murati's startup just shipped its first model, too.

DerekWait, seriously? I thought they were still heads down building.

LaurenThe Wall Street Journal reported Wednesday she's trying to loosen the grip the big labs have on this whole space. full space.

DerekOkay, that's a legitimately interesting move. Former OpenAI CTO comes out swinging with a real product.

LaurenAnd it lands same week OpenAI is explaining a breach and Moonshot crowning itself the new frontier.

DerekIt's a chaotic week to be an AI lab, basically.

LaurenWhen is it not, though?

DerekTrue, true—but step back. Every one of these stories-the breaches, the Kimi hype, Muratis's launch-they're all pointing at the same underlying fight.

LaurenWhich is?

DerekWho actually controls the next generation of this technology, and who gets to define what rivals GPT even means.

LaurenAnd that fight isn't just happening in labs, it's happening in boardrooms writing enormous checks.

DerekI see where you're going with this.

LaurenBecause if a breach in a scrappy Chinese startup can shake investor confidence in the big names this much?

DerekImagine what it does to the money chasing the alternatives.

LaurenExactly. So who's actually cashing in on all this drama? So follow the money and it lands somewhere wild.

DerekWild how?

LaurenJordan Novet at CNBC reported Nvidia-backed Fireworks just hit a $17.5 billion valuation.

DerekFor what exactly?

LaurenInference.--The machinery that runs AI models cheaper once they're already trained. Novet's piece noted Fireworks used to lean on revenue from Cursor, the coding startup, but they've diversified as more companies chase lower cost models.

DerekOkay, but seventeen and a half billion for Fireworks? Where's the revenue number?

LaurenNovet's report didn't publish one, and that's the pattern this week. That's a nice story, but show me the numbers.

DerekI don't have them; nobody does yet.

LaurenGREAT PATTERN.

DerekWait, it gets weirder.

LaurenThe Wall Street Journal says chip startup Etched is in talks for a twenty billion dollar valuation.

DerekEtched, the ones betting everything on one chip architecture?

LaurenThat's the one. Think of it like General Chip's or Swiss Army knives: Etched's play is a scalpel built for one surgery.

DerekAnd if the surgery changes?

LaurenThen the scalpel's worthless-that's the risk investors are pricing at- at twenty billion dollars.

DerekAnd the Wall Street Journal made clear that number still just talk.

LaurenRight, talk stage, not signed.

DerekTalks stage are different animals.

LaurenNow here's one that broke my brain a little. Go!

DerekColossal Biosciences, the de extinction company, is reportedly eyeing twenty to thirty billion dollars.

LaurenWait, wait,

Speaker 3wait.

LaurenThe woolly mammoth people?

DerekThe woolly mammoth people. Marina Temkins piece for TechCrunch says they could double or triple their last valuation. That's not a chip company or an AI lab; that's a biotech moonshot in the same range as OpenAI's rivals.

LaurenWhich says something about where money's willing to go right now—anywhere with a big enough story.

DerekOr a big enough mammoth.

LaurenOr a big enough mammoth.

DerekFair. The one I actually like? CuspAI raised four hundred and fifty million dollars.

LaurenFour?

DerekAI for material

Laurenscience—discovering new molecules computationally—Maria Deutscher covered it for SiliconANGLE and Reuters confirmed backers include the UK government and Jeff Bezos personally.

DerekWow! Government money and Bezos money in one round—unusual pairing!

LaurenReuters framed it as the UK planting a flag in AI-driven material discovery instead of chasing another chatbot—think batteries, carbon capture materials. Chemistry you can't fake with a demo.

DerekCompare that to thirty billion for reviving mammoths.

LaurenExactly the contrast!

DerekThat's the valuation this week I don't need convincing on; you get an actual material at the other end.

LaurenVersus Fireworks inference math or Colossal's mammoth math.

DerekMoney chasing hype is one story tonight; money chasing something you can hold in your hand is another.

LaurenSpeaking of stories with less hype and more mess— Lawsuits are up next.

DerekOh, I've been waiting for this part. But there's another side to this AI money story way less flattering. Let's talk lawsuits.

Speaker 4Oh, the lawsuits. Sony's back at it, right?

DerekYeah, Wendy Lee at the L.A. Times reported Sony Music just filed another lawsuit against Udio, this time claiming over 30,000 recordings got ripped off.

Speaker 430,000. That's not a cease and desist. That's a spreadsheet.

DerekAnd get this: Sony is the last major label that never signed a licensing deal with Udio, according to Engadget's Mariella Moon. Everyone else made peace. Sony's still swinging.

Speaker 4Meanwhile, one AI music startup just handed labels a much muddier problem.

DerekWait, the hack?

Speaker 4The hack, Suno. Someone used an employee's credentials to break in, and according to TechCrunch's Amanda Silberling, In the handling the leaked code shows exactly how Suno trained its model.

DerekLet me guess, YouTube?

Speaker 4YouTube and Deezer. Music Business Worldwide's Mandy Dalugdug says the code points to years of scraped audio, and Pitchfork's Alex Suskind adds lyrics scraped from Genius, too.

DerekThe irony: an AI company accused of stealing music gets hacked and the hack proves the stealing.

LaurenPoetic, caught red handed by her own break in.

DerekAnd if Suno's really been doing what everyone accused it of, that undercuts every "we licensed everything" defense these companies lean on.

LaurenUdio's already fighting Sony without a hack, now Suno's fighting with leaked receipts.

DerekFeels like the labels finally have leverage instead of just headlines.

LaurenFor sure. Evidence beats a strongly worded lawsuit every time.

DerekMakes you wonder if investors get nervous when the target of a of a copyright suit is also the target of a breach.

Speaker 4Probably. Funding conversations get a lot harder when your training data becomes evidence in someone else's case.

DerekAnd it has to make future negotiations weirder. Why sign a licensing deal if your competitor just got proof they didn't need to?

Speaker 4Exactly the wrong incentive: Cut corners, and if you get caught, you're already in court anyway.

DerekSongwriters aren't just mad about money, they're mad about becoming training data nobody asked them about. about.

Speaker 4I mean, that's the part that never makes the press release.

DerekNot a good week to be an AI music startup's legal team.

Speaker 4Not even a good day, honestly.

DerekOkay, switching gears because I need less courtroom, more actual hardware.

Speaker 4Perfect timing, because apparently a Formula One engineer got bored racing cars and started building robots instead.

DerekWait, seriously?

Speaker 4Patience, we're getting there.

Speaker 5you

DerekSwitching gears: actual robots now.

Speaker 4Thank God something with a body attached.

DerekRight? So there's this humanoid robotic startup, humanoid, that just closed a Series A. Reuters says one hundred and fifty two million, valuing the company at one point three five billion.

Speaker 4Wait, a Series A at that valuation?

DerekYeah, yeah. Investors are betting on a robot that can actually walk a warehouse floor and pick

Speaker 3up stuff.

DerekYou'd pick things up. I mean warehouses have had robots for years; what makes this one worth one point three five billion?

LaurenBipedal; full humanoid form, not just an arm bolted to a rail.

DerekAmbitious.

LaurenOkay, so get this, I've got an even better one-former Red Bull Formula One engineer.

DerekWait, from the racing team?

Speaker 4The actual pit lane. Business Insider's Rya Jetha reported he got bored designing race cars and started a robotics company called Micro AI.

DerekBored with Formula One.

Speaker 4I know. But he's putting AI-powered robots into factories, and he just raised fifty five million dollars to do it. Wow!

DerekThat's a jump from split second pit stops to slow, careful factory work.

Speaker 4Different kind of precision, same obsession with shaving off seconds.

LaurenSeconds.

DerekOK, but here's what I actually want to talk about: the people behind these companies. Andreessen Horowitz's Zabie Elmgren wrote this piece on the seven hires a hardware startup has to get right.

Speaker 4And?

DerekHer point is the engineering team can build something incredible, but without the right ops and supply chain, the whole thing falls apart before it ships. I live through hire number four myself. Ops lead brought in way too late. It nearly sank the whole roll out!

LaurenWhich tracks with what we're seeing: raising fifty five million or a hundred and fifty two million is the easy headline. Hiring the person who keeps your factory line running is the boring part nobody writes about.

DerekLet's get real for a second: every hardware founder I've talked to underestimates that hire. They think it's all engineers. Then six months in, they're missing a manufacturing lead.

Speaker 4Mm hmm.

LaurenSo between humanoid robots- That's ex F1 engineers and A16z basically handing out hiring cheat sheets.

DerekIn China, physical AI is having a moment.

LaurenA real one, not just chatbots pretending to be smart.

DerekCareful, don't start a fight with the language models.

LaurenToo late. Anyway, next year I want to see one of these humanoids working an actual shift somewhere, not a demo stage, a warehouse.

DerekA full shift, no fainting, no falling over. Over.

LaurenBar's low, but I'm optimistic.

DerekOkay, so if there's one thing to take away today, it's that

More episodes

Keep listening

Other episodes of Tech Insider Weekly, newest first.

All episodes of Tech Insider Weekly

Sources

Where this came from

20 reports behind the episode. Every one of them opens where it was published.