# Anthropic IPO Panic and AI Infrastructure Debt Risks

**Podcast:** Big Technology Podcast
**Published:** 2026-10-16

## Transcript

Anthropic is growing fast, but its growth rate is slowing, leading some investors to panic ahead of its IPO.
Was there a spy at OpenAI?
And it's time to end food slop, or is it?
That's coming up on a Big Technology Podcast Friday edition right after this.
How do you price intelligence as a service?
How do you charge for delivering end-to-end work, not just access?
Those questions remain front and center for AI founders and operators.
And getting to the right fit business model...
Whether that's selling usage credits, outcomes, seats, or a hybrid combination requires constant iteration.
Chargebee is the billing and monetization infrastructure that powers just that for ambitious, fast-scaling AI businesses.
Visit chargebee.com slash Alex to learn more.
That's C-H-A-R-G-E-B-E-E dot com slash Alex.
Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional cool-headed...
and nuanced fashion.
We have a great show for you today.
We have the latest Anthropic numbers ahead of its IPO.
Finally, some 2026 revenue.
It looks good, but is it good enough to avoid a full blown panic as the market digests the numbers?
We're also going to talk about the reported incident of corporate espionage within OpenAI, what it means and whether there's more afoot.
And then finally, Ranjan and I are going to debate whether all these AI-generated images on restaurant menus and signage are good for humanity or the end of us.
And that will be fun.
Wait for that at the end.
So joining us as always on Fridays is Ranjan Roy of Margins.
Ranjan, it's good to see you.
Welcome back.
Good to see you.
I believe AI-generated food slop images represent the American dream better than anything else.
But we'll get to that.
Oh, my God.
Please, please, please do not.
Do not.
seriously back this belief.
I can't stand the AI signs, but okay, we'll hold the debate till the end because right now, finally, what you've been asking for, the Anthropic 2026 numbers are here.
They're very impressive, but I think expectations have been set so high that no matter what Anthropic showed, potentially, there would have been some panic and we are indeed seeing some panic today.
So let me read the story.
This is from the information.
Anthropic's rapid but slowing growth could weigh on its IPO valuation.
Anthropic reported rapid but slowing revenue growth.
In a meeting with potential investors, the company said revenue hit $16.8 billion in the third quarter.
That was a jump of about 45% from the second quarter.
The figure was more than 3.6 times the company's revenue for all of 2025.
However, and now there's a big but here, but it also represents a slower growth rate.
Then Anthropics' second quarter, when fevered demand for its Claude Code product drove up revenue by more than 140% from the first three months of the year.
So here's the interesting thing.
So Anthropics' revenue, we know it was in the $4 billion range for all of 2025.
It grew 140%.
from the first quarter of the second quarter this year, but then only grew 45% from the second quarter to the third quarter, reaching $16.8 billion.
That's not enough for a lot of people.
And we're already starting to see some of these tweets come through saying it's over, some worry that Anthropik's growth rate is slowing, which of course, like law of large numbers, it was eventually going to.
And the party seems like it might be...
not over, but more muted than initially anticipated.
People are saying that this shouldn't be a $2 trillion IPO.
It's probably going to be closer to $1.5 trillion.
And of course, there could be a ripple effect to all of those that were expecting Anthropic to be bigger.
So what's your read on the numbers, Rajan?
My read on the numbers is release the damn S1.
This is getting so frustrating.
Again, and we're going to get into some confusion around a Reuters report, but overall, it's just a Rorschach test right now.
I mean, everything we see, it's clear.
I mean, 45% off of that base is incredible growth, but it is...
When you start to extrapolate it out, they would have to re-accelerate growth to get to that $100 billion number, even an $80 billion number.
Basically, if they take, I think it was around a 7% month-on-month growth, if you extrapolate that out, they get to about $75 billion in revenue.
Again, any of these numbers on their own in any other context would be remarkable, fantastic, unprecedented.
to get to that $2 trillion valuation, it's not enough.
But then, I mean, there's a lot of questions around it.
The distillation is fascinating when it's 5% to 7% of total revenue.
Okay, you've got to introduce that.
Talk about the distillation.
Okay, okay.
So distillation, again, external companies basically accessing Anthropic to ask...
billions of questions probably in order to actually understand what the answers are and use that to train their own models.
Now, which companies exactly are doing that is not official, but lots of companies are.
And I think that's probably why we saw an equivalent explosion in terms of open weight models from a lot of companies.
So when you think about 5% to 7% of revenue, it's interesting how- Wait, wait.
So just to say, the anthropic- told investors that distillation is 5% to 7% of its revenue.
And it's working to rein in that distillation, right?
So companies trying to use its answers to train their own models.
It's trying to rein that in.
And so there's going to be a revenue slowdown from that.
I think that's important to state.
Yes, yes.
There will definitely be some type of revenue slow.
That revenue goes away because Anthropic is cracking down on that kind of behavior.
What's interesting to me is...
it still fits so perfectly that 140% quarter one to quarter two growth.
I've said this over and over on the show.
Like, I mean, I felt it.
I saw it being in the industry.
Like it just, those spring months, that's.
all anyone could talk about.
Claude code completely exploded.
Every CIO is bragging about how much money they're spending.
But also then you have this extra headwind of foreign companies actually, or maybe even domestic companies, distilling anthropics models.
So that definitely contributed to that.
absolutely explosive growth.
If it's 5% to 7% of total revenue, it was probably concentrated during that period, which you can imagine actually for that growth period, it contributes even more.
So I think it is a really interesting dynamic.
I don't think it fundamentally changes the story.
Maybe it does.
I mean, that's still a pretty dramatic impact overall.
It's not marginal.
But it also, when you think back, That growth is what got them all the hype, additional funding that they can then pour into more marketing and training and model building.
So I think that is a pretty important dynamic that can change the story.
Okay.
And I think this is important to take a step back and say what we're saying is, is this growth slowdown?
Is it a story of the numbers being too big?
Or is it a cool down from this period of everybody vibe coding, everybody cloud coding, everybody cloud working, co-working?
And what you're saying is...
Yeah, we are.
There was this surge of interest that has cooled off.
And so the growth, even though it's impressive and the numbers are very big.
Again, I don't want to I don't want to diminish the fact that a company that made four billion in twenty twenty five made sixteen point eight billion in the third quarter quarter.
Right.
So three point six X the entire company's revenue in one quarter.
But the thing is, it's all about.
expectations and where people sit in terms of their expectation.
I think there was this expectation that that moment, the Claude Cohn moment in the second quarter, was effectively a permanent moment or that was not a spike, that was a moment along the exponential.
But the numbers are showing that maybe it wasn't.
And, you know, there were people who hit me in the comments when we talked about Anthropics numbers, you know, initially about this Aptopia data that I brought into the conversation saying it's not representative.
And, you know, of course, we talked about it also.
We caveated it also.
Right.
But I'm going to just read it again.
Claude's monthly U.S.
downloads went from nearly 670,000 in January to a peak of 7.42 million in March.
But it dropped to 3.26 million.
in September's.
Yes, it's an app.
Yes, it's a proxy.
But we take these numbers as signals, right?
Not the full story.
And it turns out that signal might have actually been telling you something.
I actually, I think that's fair.
It's a signal.
I was going to push back and I still think, again, this is a company that last year, 25% of the revenue came from two customers.
Like everyone kind of knows like Claude is not everyday use.
It's Hardcore, concentrated use, very expensive use, a lot of enterprise use.
So app downloads, I generally wouldn't think, but I think that's fair, that it is yet another signal.
I think the most important part of this conversation, again, going back to that, I just keep seeing the Rorschach test side of this, of like everyone who has been bearish.
This is a clear signal that it's not going to go well and you're not getting your $2 trillion valuation.
Everyone who has been bullish is like, are you kidding me?
45% growth at that base?
That is, again, unprecedented.
So to me, the most important thing is just give us the actual numbers.
Again, do you realize how crazy this process is?
Like in terms of unprecedented, to have a large institutional investor meeting.
where you disclose some financials, that is not part of the official IPO process.
This is not a traditional testing the waters process, where it is assumed information is going to get distributed in some kind of way.
Like, this is just not healthy.
And then that literally the entire global economy rests on this.
South Korea's stock market is down 8% on this kind of news.
I don't know.
This has been killing me.
I actually saw Imran Khan, who was a very longtime investment banker, chief business officer at Snap for a while.
He actually said, he's like, I still don't understand the point of a large institutional investor meeting while Anthropic is in confidential filing.
And he even says, given their S1 leak to the news media, I've never seen another in my 26-year career the company should have been more sensitive.
So honestly, that's the more frustrating part of this whole thing for me.
Right.
And for our non-financial listeners and viewers, the S-1 is the document that a company files with the SEC with all the numbers ahead of going public.
But Ranjan, let me take a different angle here.
which is that maybe the S-1 doesn't matter so much and maybe what matters more is the expectations that the company is setting again through the press or through leaks, right?
Because we had talked about it, the Times reported the expectation was Anthropic was going to do 100 million in ARR or sorry, annualized revenue by the end of the year, right?
So that is the number that everybody has been looking toward and that is where their expectations are set.
And so What we saw with the numbers that have leaked is, and I think you put this very well, is that unless the growth rate accelerates substantially, getting to that 100 billion, sorry, if I said million, I meant billion, 100 billion ARR is going to take a market acceleration and growth rate of revenue as opposed to the deceleration that we've seen.
Well, but this is where it gets so confusing.
Because $100 billion, it's not $100 billion ARR, right?
Or is it?
Is the expectation $100 billion ARR, meaning in December you're showing, what, like $8.5 billion in a month or something like that?
Versus what is the expectation for the total revenue for 2026?
That's the...
more important thing.
And that's where it gets so muddied because like they're not getting a hundred billion dollars revenue in 2026.
There's no, I don't think anyone's talking about that, but it just, I don't know that there was even Reuters reported and this, and listeners might sense my frustration at all of this, but.
Oh, we can sense it.
Yeah.
Reuters had this reporting.
that then said there was $110 billion ARR from, like, investors who attended the event.
Then they clarified.
I mean, so obviously then everyone who is bullish is like, wait, $110 billion ARR is projected.
Then the reporter clarified that that was actually from sell-side analysts who did their own calculations based on guidance.
Like, this is, again.
The economy, entire nation's stock markets are resting on these leaks.
And we're all just going in so haphazardly into this.
Just stay quiet and then release the S1 as every other company has done.
Can I ask you a question?
Is there an argument to be made that basically no matter what they release, this sort of volatile vibes based AI trade?
is going to remain that until all these companies get bigger.
I mean, if you think about it, you know, we talk, I mean, even if you look at the numbers coming out now, you know, 16.8 billion, 16.8 billion in a quarter, they're still trying to get to a $2 trillion or $1.5 trillion IPO.
And, you know, so much of this is based on a story.
It's based on this idea that, you know, remember, they haven't even figured out the products yet.
How many product shifts have we seen in the past year?
Chatbot.
vibe coding, co-work, agents, right?
So the final form of this technology is still not found.
And so to expect any sensible trade, right?
So we're talking about the IPO panic.
To expect there to be a lack of panic and a lack of these big shifts whenever you see a number.
My question is basically, Isn't it the case that no matter how much financial information we're going to see, that is just going to be a feature of this until it either gets to maturity and achieves its potential or doesn't?
I think that's a good question.
I think anything that gives more of a reflection on what are the economics of the business.
Because, again, they said, I mean, sorry, they didn't say it was reported.
And after that Reuters mishap, I want to be extra cautious on this.
That they were adjusted profitable, whatever that can mean.
And we know, I'm guessing, probably Anthropic does not remove partner expenses from partner marketplace expenses like OpenAI does, which caused confusion with the FT recently.
Whether they're going to include training costs.
as an R&D expense or as part of the cost of goods sold.
That's going to be another element of this that will fundamentally change that story.
They said they were cash flow positive.
To me, I think you're right, though, that we don't know what the end product is.
We don't even know what the end business is.
Is it going to be frontier models that are expensive driving the majority of their business?
Is it going to be the cheaper models?
Are they going to compete against the open weight models?
We don't know the product.
We don't know where the actual business model will go.
So yes, I think I agree that it is all going to be vibe Z, but some kind of disclosure should give us a better picture of that.
Hopefully.
Maybe it won't.
Well, the question is, when is the IPO actually going to happen now?
because there was some talk that it was going to happen in October.
Then people started talking about it's November.
Now I'm starting to see in stories.
Well, that's subject to change.
So I don't know, Ranjan, you might be waiting for a while.
I was going to say, you know, this stuff should be imminent.
But I think this conversation with these bankers suggests that we might be waiting, you know, much longer.
Watch the S1 drop as soon as I say this.
But it doesn't seem like they're rushing in, especially now that the market is really showing this sort of uncertainty about the AI trade.
Of course, we saw the market movement that we talked about last week in terms of OpenAI getting punished for a FT report that talked about its revenue, but actually the revenue didn't change.
It was just some people measuring it differently, according to reports.
But now we're also seeing this uncertainty around the neoclouds.
And not much has changed on the ground this week in terms of companies like Corweave and Ebius and Iren.
But the market is punishing them pretty intensely right now.
And it's hard to really find the compelling event.
So let me just read a little bit of the story here.
This is from CNBC.
Traders pile into bubble puts on Corweave as Neocloud's stocks plunge.
Gotta love the term.
bubble puts.
I've never heard that before, but here we go.
Here's the story.
A key corner of the artificial intelligence trade is taking a beating and traders are betting that it could get even worse.
Shares of neocloud companies plunged after debt yields for Nvidia-backed core.
We've continued to reach new highs.
Meanwhile, social capital founder Chamath Paliapatiya raised questions about a private credit bubble alongside a Screen grab showing Corweave's bonds hitting a 15% yield.
Elon Musk replied, yikes, Corweave stock plunged 15%.
Nebius Group and ERIN, popular trades connected to the data center buildouts also fell as much as 10%.
And now options volume in Corweave is surging to more than seven times the 30-day average.
It had the most bearish action of all of the neoclouds.
200,000 puts compared to 106,000 calls, which means that they are basically two to one betting that the stock has even further to drop.
So Ranjan, I want to turn this to you.
Why do you think this is happening?
People are saying that its interest rates are rising, which means their debt costs will go up, which means maybe the project is not really sustainable.
What do you think?
And is this another example of how the vibes are poor around these companies and we're starting to see the impact?
I think on the NeoCloud side, the vibes are even poorer.
Because again, Anthropic still showed us they are building a gigantic business.
Like again, as we've recognized over and over, and it's fair, like $16.8 billion in a quarter when you're four and a half last year is insane.
But on the NeoCloud side, you have all these companies.
I actually...
My favorite story from this week, which is crazy to me that it's not bigger, is there's a company, Firmus, in Australia.
They were supposed to, NeoCloud, NVIDIA and others are big investors in it.
They're supposed to go IPO at a $44 billion valuation.
There was actually these really interesting dynamics around it where their investors had such kind of...
favorable investing terms that basically they were able to kind of stop the IPO and pull out because if it wasn't going to reach a certain valuation, they would have to be issued a lot more shares and receive financial benefits.
This was a company that was founded in 29.
Like I had never, have you heard of firmness in the past or paid attention?
I can't say I have.
I have not either.
I mean, I'm like, perhaps that is out of ignorance, but like, this is a company that was a neocloud that's.
was going to go public in Australia at $44 billion.
This was also a company that in 2019 was founded as a Bitcoin mining company, which CoreWeave was as well, and I understand the pivot to AI infrastructure.
It was also founded by two people, one of whom was in jail for insider trading for a year in 2015.
That's why I'm building my neoclouds.
No, no, his co-founder is his cousin.
who had no previous technology experience other than running a family lingerie business and no hate on coming from a dorm in Victoria's Secret.
No hate on, yeah, exactly, a tech-focused lingerie business.
But you know what?
I still recognize I should not be building an AI cloud infrastructure business.
I say it, and I work in AI.
Don't trust me with $15 billion of debt to build a data center.
No one ever will.
I mean, sorry, sorry.
I'm just, you know, preventing future opportunity for myself.
But the main thing is, like, it's such a ridiculous story and it's not even the headline.
Like, again.
former convicted insider trader who spent time in jail, started a business with their cousin, Bitcoin mining, has not actually delivered any infrastructure yet, is just taking on more and more debt, was trying to turn that into an IPO.
Like, this is, the neocloud side of things, I think, is dangerous.
And I think the market is definitely starting to recognize it.
Right.
And you know what?
I said maybe vibes-based, but maybe there's a non-vives-based part of this that I want to run by you, given your financial background, which is that we have seen borrowing costs go up.
Interest rates are on the rise as inflation fears seem to be settling in.
And we have more predicted rate raises in the future.
And we've talked about how these companies really need a lot of debt in order to grow.
And when their debt becomes more expensive, I wonder if that changes their ability to provide the computing because that's a pretty big hit in their margins, don't you think?
So what do you think about this?
So I actually love – this is like the one part of the entire financial market infrastructure that appears to be efficient at the moment.
The massive amount of debt that is being raised increases the demand for capital, which should increase the cost of capital, which is the interest rate, and interest rates are rising.
Now, when we see the 10-year, I think it hit like 520 maybe before, 530, 5.3%, and it's back down around 5.05 as we speak.
That is what's supposed to happen.
The more people that are trying to raise debt, it should increase the cost of that debt.
And it's happening, and it's funny because that's not good for them in the long run, but that's also just a reminder that capital is finite.
A lot of this has run off the idea that capital is infinite, and there is going to be an increase in cost to it.
And debt, again, core weave, it's pretty crazy because the bonds that...
They're issued at 9.625%.
They're currently trading at a 15% or I think it hit a 16% yield.
Corweave's bond.
So Corweave is paying 15 cents on every dollar it borrows right now.
Well, that's what investors are trading the bonds at, assuming that that's what they're worth because of the implied default risk.
Oh, okay.
So they're not, because they already issued the bond.
So they already got that money at 9.6%.
But now the market is assuming that there is risk of default implied in that or not like some kind of default.
So for me to buy that bond, I'm going to charge you more.
You should charge me a higher interest rate or a higher yield.
Basically, it's not good.
It's effectively distressed territory.
I mean, I think, again, I think, you know what, this makes me happy.
We got some efficient markets left.
You know what, the bond market is not even as liquid as the equity market, but even the bond market, and a lot of people always say this, that like the bond market is ahead and smarter and they clearly are trading like things are supposed to trade.
So there's some hope for the market yet.
Get Ranjan talking about interest rates, which is actually, Ranjan, how we began our conversations like four years ago.
You got a happy man on the other end here.
I know.
I never, I didn't like when there was zero interest rates that caused some issues.
And I'm actually saying now rising interest rates.
And I say that and recognizing it's not good in many ways, but at least in this case, it just shows someone is paying some attention and trying to be rational.
Look, before we move on, I just want you to sort of.
Explain again the 15% yield on the bond, who's paying who and for what.
That's what the bond is trading at in a secondary market.
So the most price is inverse to yield.
So bonds are, it's actually, there's a price associated with that.
So if it was like 97.5 before.
The higher the yield shows, it's actually trading.
There's like actual 96, 95, whatever it is.
So it's actually, the way to think about the bond market is those higher yields, it basically means someone is actually paying a lower price for that asset.
So it's just showing people are paying lower and lower prices.
It is kind of weird and counterintuitive, but it's how everyone kind of looks at it.
But basically, price is going down.
whenever you see yields going up.
And it should be- So the Corweave bonds that were issued to pay for the infrastructure that they're building, if Corweave sold you those bonds and you want to resell them, then you have to pay a premium or you're selling them actually for less than you paid.
You're selling them for less than you paid.
That's it.
You're taking a haircut on core weave debt right now.
It's basically that.
And so the higher the yield, the lower the price you're going to get.
Exactly.
Yeah.
It is funny now that, I mean, it's literally, with equities, it's always simpler.
If it's at 100 and then it's trading at 90, we all know the price went down.
But that is effectively what it is.
Okay.
Let's end this segment.
Do you think this is an overreaction and Anthropic is just going to be fine?
My take on all this is there's so much vested interest that they go out successfully and energy.
I still think they're going to go out okay.
I don't know.
I think it's still going to take a lot to really derail this.
But I don't know.
What do you think?
You know, we've talked about all the risks that Anthropic has.
Honestly, we've talked about the fact that the frontier is commoditizing to some degree.
We've talked about the fact that there is competition now on the enterprise side with a place like OpenAI.
We've talked about how some of the big customers are pulling back from all the token spending, customers like Meta and Microsoft.
And we've also talked about how the enthusiasm around the cloud code and the cloud co-work moment, which led to a lot of revenue for Anthropic, has dulled a bit.
So, you know.
I still think this is a strong company, but to think that this is an inevitable company or that it will sustain the success it's had to the same degree in an industry where we've seen companies, you know, take the lead and then go to the back of the pack in a matter of months.
I don't think you can say that necessarily Anthropic will be able to sustain it.
And so with so much on the line, I don't think it's really an overreaction at all.
I think, yeah, I was just thinking about.
This is where Elon, his greatest gift, I still, is the storytelling and the timing.
Because to go out at a trillion plus with $19 billion in revenue, which is what Anthropic made in the last quarter.
but he did it early.
And also, like, Mars and stuff is such a better story where there's no competition in the future of humanoid robots or whatever else it is versus Anthropic is in the biggest dogfight imaginable, and it's the most competitive environment imaginable.
So, yeah, there's a lot of risk, but this is, I hope there's some resolution to this that we just know or don't know in the next month or two, but.
I don't know.
Do you think they go up by the end of the year?
Definitely.
I definitely do.
I was going to say when you were saying, I hope there's some resolution.
Ranjan, we will see that resolution for sure.
All right.
You've heard it here.
You heard it here.
The Cantrowitz guarantee.
The Cantrowitz guarantee.
I'll put the stamp on it.
All right.
We got to talk about this reported case of espionage within OpenAI, an employee.
took a recording device and put it in an executive conference room.
We'll talk a little bit about what happened, the implications, and whether there might be more such incidents within companies in Silicon Valley.
That's coming up right after this.
This episode is brought to you by Intelligence AI.
Is your investment in AI coding tools helping your team ship better software?
Intelligence connects your AI spending to code quality and engineering outcomes from individual sessions and pull requests to entire projects.
You can see where AI saves time, where it creates rework, and how to improve.
and its model router selects the right model for each task based on quality, speed, and cost.
See how Intelligence AI helps enterprises optimize AI-powered delivery with connected visibility into cost, quality, and outcomes at intelligence.ai.
Everyone's talking about AI, but there's a big difference between running a promising pilot and actually making AI work across an enterprise.
That's why we're launching the Big Technology Guide To series presented by DXC Technology, available on YouTube.
Throughout the series, we're tackling some of the biggest questions facing businesses today.
How do you take enterprise AI from pilot to production?
How do you navigate an increasingly complex cybersecurity landscape?
What's the real opportunity with AI agents?
And how do you calculate the return on your AI investments?
On my Alex Cantor's YouTube channel, you can hear from experts like DxT Technologies Chief Digital Information Officer Russell Jukes.
When you ask a question to somebody in your company, do you trust them?
And what made you trust them?
Do you trust but verify every single time?
Russell and I discussed the challenges and opportunities of building production-grade AI and what it really takes to move beyond the hype.
Check out the Big Technology Guide To series presented by DXC Technology on my Alex Kantrowicz YouTube channel today.
And we're back here on Big Technology Podcast Friday edition, talking with Ranjan Roy of Margins.
Ranjan, very interesting story in Silicon Valley this week coming out of OpenAI.
This is from the Wall Street Journal.
OpenAI engineer allegedly hid toilet paper wrap mic in a conference room.
As someone who is proficient in the use of microphones, I will tell you, you always want to wrap them in toilet paper to hide them.
Should we do that?
Should that be the new thing?
It's the new green screen.
All right, let's talk about it.
Here's the story.
An OpenAI engineer visited a familiar conference room in one of the company's San Francisco office buildings last week and peered behind a television set mounted to a wall.
Abdulrahman Al-Fozan had spent weeks scoping out the room, which could only be reserved by executives at the artificial intelligence company and their assistants.
The conference room was often used for leadership meetings, briefings with government officials and business partners.
as well as top secret research and product discussions.
Al-Falzan, who had worked on OpenAI's product team for more than a year, planned to intercept what was said within its walls, the Justice Department said.
He was charged with using or endeavoring to use a device to intercept oral communications.
He's a 35-year-old Saudi, and he's accused of rapping.
the thumb-sized circuit board and microphone and toilet paper and inserting the device into the USB port on the back of the wall-mounted TV in late September.
He's been arrested.
So I just want to say we don't fully know.
Look, the speculation has been that this guy was working on behalf of the Saudi government.
We don't know that.
I think it'd be irresponsible to speculate that that's the case.
It could have just been a curious employee who wanted to know what was going on in the executive room.
This stuff happens.
But ultimately, I think it has...
awakened people to this idea that, you know, regardless of whether he was working on, you know, or some entity connected with the Saudi government, that this stuff is quite possible within companies in Silicon Valley, whether it's individuals or whether it's nations.
So Ranjan, what do you think about this story?
And what was your reaction when you saw it?
I mean, this was a fascinating story.
I agree, though, that like, what the geopolitical implications are were...
are very unclear right now.
And whether we ever find out or not, I'm hoping we learn a bit more.
But I got to say, I was kind of like really interested in what he actually did because someone had actually, there was reporting that the device was a Seed, Seed with three E's is the name of the company, Xiao ESP32 S3 Sense little chipboard.
It's like $8.
And then you add a little microphone to it.
My favorite part of the story is part of how they caught him is he actually was asking ChatGPT how to build this, which is amazing that he actually asked.
Use Gemini there, man.
Or maybe Gemini can't do it.
Allegedly.
He allegedly did this.
Sorry, allegedly.
But just because I'll admit, like, when I went on that company's website and started looking at these devices, I literally was like, oh, this is interesting.
This would be a fun.
weekend hacky project to do just to, again, run actual recording off this tiny chip.
But I don't know.
I mean, this is crazy to wonder how much of this is going on at these companies.
Again, where secrecy is paramount, where they're moving so fast.
I mean, the hiring numbers, the turnover, like all this is so crazy that, and it is a reminder, the advance in technology that you, don't have to be like a spy to be able to put together something like that.
You can just use a chatbot is pretty amazing too.
Yeah, I did love the there was a tweet from OpenAI.
We used our models to understand the device and where it had come from, identifying that a specific employee had planned it a few days earlier.
And then somebody quote tweeted that with image of Dario Amode in a wig and a fake mustache and a big smile.
My favorite part of that tweet from OpenAI is like, are they saying, when they say we used our models to understand, do they just take a picture and it's like, yo, Chachi BT, what this?
I mean, I do that all the time with all different AIs.
use it to try to understand some kind of device or some kind of electronics and like we used our models is it literally hey chat what am I looking at here you are looking at the and this was the person who did it because I have their chat GPT history and I will tell you all about it right now Yeah, I mean, this is one of the things when the details of this came out, it's like you didn't really need a sophisticated, you didn't need Mythos or you didn't need Asper 6.1 to figure out what was going on.
In fact, like, and again, this is all reported and alleged.
This is from the Wall Street Journal story about what happened.
The small device had been in place for about two days.
OpenAI reviewed footage from its surveillance cameras and saw that El Fosan had been near the conference room where it had been hidden several times before the bug was found, despite having no reason to access the room.
In the weeks beforehand, Prosecutors allege Alfosan had used an AI tool to query meeting rooms and locations of leadership and exec meetings.
The federal complaint also says a chatbot associated with a spouse asked multiple questions related to audio recording, remotely accessing a device, and encrypting a data card.
Last week, prosecutors say that he left a bag unattended in a different part of OpenAI's.
offices and inside the bag they discovered his wallet and a similar recording device to the one they found in the executive conference room he was arrested on Friday it's like man you're gonna leave your bag with your wallet and a recording device like you might as well like on the toilet paper write your name I mean but that's what that's what is I love about this story and it's so fascinating that like The actual technology involved would have in the past required, what was the James Bond guy who always has the Q or whatever?
You know what I'm talking about, right?
Yeah, that guy.
Not a big Bond guy, but you know what I'm talking about, listeners.
The person who presents James Bond with all the gadgets.
That in the past, this kind of like mini listening device that's actually able to record over a long period of time.
And like that would have been.
really difficult or required like being like a spy and now it's just someone on chat GPT asking these questions and it's a reminder that they're not a spy and they're going to make some really dumb mistakes when they're doing this so good and bad Yeah, I mean, the District of California, so the U.S.
attorney, seems to indicate that this was a national security issue.
The arrest highlights our relentless efforts in partnership with the FBI to identify and prosecute those who seek to steal our technology.
For us, it's a matter of national security.
I don't think you would say that unless you thought that there was something, you know, international going on.
But the lawyer representing this guy, El Fosan, said his client was charged, was...
was charged with violating a federal recording statute, not stealing trade secrets or working with a foreign agent and asked everyone to withhold judgment until the full effects become known.
I think that's a fair request.
But you know what this thing has done is it's woken up people's imagination to like how many countries, you know, whether or not this guy was doing this, you know, in cahoots with the Saudi government.
It's woken up.
people to like how many countries have people placed within these labs.
And I remember I wrote this story a number of years ago about actually two people, two Saudi nationals within Twitter.
when it was called Twitter at the time.
I remember that, yeah.
Who had, yes, who had, I'll just give the outlines of it.
They had worked with somebody associated with the Saudi royal family.
And one of them, I think, well, anyway, they had allegedly worked with the Saudi royal family.
This is according to like the prosecutor's documents to like access the information of Saudi dissidents and pass that along, which of course, like if you have an anonymized account and you unmask them to the Saudi government, that's not good for the.
It's not going to be a good time for the protester, you know, put it that way.
And I remember in my reporting for this story, I had spoken with somebody within one of these companies and they were like, without any hesitation, that there are nation state actors, you know, so people with ties to intelligence agencies inside every company in Silicon Valley doing stuff.
like this.
And it does make you wonder in this, in such a high stakes case, or such a high stakes technology, how much of that stuff is going on everywhere within the labs, within the tech giants.
Interestingly, the recording device, you know, in this case, and again, we don't know if it's tied to national intelligence, but it wasn't there to exfiltrate the weights.
It was there seemingly to listen to what was going on.
and high stakes conversation among leadership about research, future research and, you know, other important partnerships.
Very interesting case.
No, I know.
I think it was about understanding the strategy direction concerns, but I keep wishing that they set this entire thing up and then in reality it was a bunch of like open eye executives.
like worried about Muse talking about dots, looking at Alexander Wang's Twitter feed and being like, he sexualized the labubu.
He's shitposting his way to a 20% increase in the meta.
He's calling himself an Asian lesbian, which he did today.
And it was kind of a funny, yeah.
I can't believe he put Muse's bot on main once again.
So whoever has put him up to this is sitting there waiting for this like deep.
corporate, like, what is the strategy, hypothetically.
Simon or himself wants to exfiltrate that kind of information.
And it's just a bunch of people scrolling at Alexander Wang, trying to figure out what they need to do with dots.
Yeah.
That's what I hope.
Can you imagine?
You're doing the briefing.
You've got the PowerPoint there.
You have all the OpenAI's leadership team assembled.
You don't know you're being recorded, and you put up the PowerPoint, and you say, all right, I'm going to...
turn your attention to slide five.
This is another instance of Alexander Wang sexualizing the muse LeBouBou.
We must sexualize the dot.
How can we do this with dots?
Do we have a counter?
Okay.
So interestingly enough, companies within Silicon Valley seemingly are prepared for this.
And there's another Wall Street Journal article talking a little bit about it.
It says AI companies play cat and mouse.
with secret-stealing spies.
At a San Francisco playground, a stranger tried to befriend an OpenAI employee who was there with their child.
The stranger also regularly showed up with the child and, in the process of making small talk, asked the OpenAI employee questions about their work.
It was not idle curiosity.
The stranger was a spy.
That is one of the scenarios employees of the artificial intelligence company have been presented with in attempts to raise their vigilance against attempts to steal corporate secrets.
And that was before federal authorities arrested an employee for planting a listening devices.
In training sessions, OpenAI has warned employees that outsiders, including competitors and state-sponsored actors, could try to get access to company information.
The leaders have said that this could take the form of people offering money, listening in cafes, or even trying to seduce OpenAI employees.
Employees that other AI heavyweights, including Anthropic and Google, have received similar warnings.
So I guess it's a good thing that they're getting ahead of this.
I mean, the honeypot tale is old as time.
It's still, I always- Talk about the honeypot example.
So what's a honeypot and talk a little bit about this.
Honeypots are scenarios where agents offer sex or romance to get close to a target, according to the Wall Street Journal, but also for anyone who's watched probably a James Bond style spy movie.
It's also meta strategy with Muse, but anyway.
Oh, my God.
We're all getting honey potted.
Give us your data.
Oh, dear.
Oh, my God.
Okay.
But actually, hold on.
I got to get back to this.
So one of the stories, according to the Wall Street Journal, one of the stories OpenAI tells employees to illustrate the risk.
A case where a U.S.
diplomat in Russia...
There is a foreign espyage operation where at first they sent a blonde woman to his hotel room with champagne.
He declined.
Then they sent a redhead woman.
And then they next sent a young man.
And that OpenAI warns their employees, this could happen to you.
I don't know.
I've always.
That is the official order, by the way.
Blonde, blonde, redhead, young man.
The blonde, blonde, redhead.
It's in the Honeypot Guidebook.
Husband says, OK, yep.
Yeah, it's in the Honeypot Guidebook.
But I mean, it.
Did make me like, yeah, how much of this is happening at how many levels?
There is potentially so much value to little bits of information and you can move quickly.
I was thinking about like, I mean, Apple and their tradition of like secrecy is famous for this.
But in the past, like if you find out what they're doing, that's going to change your three-year roadmap maybe.
You're not going to like massively.
change your supply chain and launch a new product.
Whereas now you get a bit of information, you can do something with it tomorrow in the next week or month.
So it has to be.
Honey pots have to be plentiful right now.
Well, listen, I mean, I don't know if I should say this or not.
Having spent some time in the San Francisco dating scene, I could see why the honey pots work.
It can be tough out there.
I'm just saying.
As a New Yorker, I'll refrain from commentary on that one.
Okay.
I mean, that's funny.
I mean, I moved to New York and I thought, okay, like, you know, I was like, maybe, I don't know.
It'll be nice to be outside of the SF dating scene.
And then I met my wife four months later in Central America.
So go figure.
One thing to note.
The honeypots are not, there's actually a big SoftBank Wall Street Journal investigation in 2020, I remember, where someone, they tried to honeypot former SoftBank President Nikesh Arora.
And he was onto it.
He didn't fall for it.
Man, Nikesh knows.
I know.
Well, actually, I mean, I'm trying to get Nikesh on the show.
So maybe I can get him to tell the full honeypot story.
You and I should do that interview together.
I feel like we bonded over this topic now.
Nikesh, I know you're here to talk about cybersecurity, but we really want to learn about honeypots, which is, I guess it is a part of cybersecurity.
And how to beat them.
Which is a cyber...
Actually, that is the best interview.
Why should you be the chairman of Palo Alto Networks?
Yeah, I didn't fall for the honeypot.
I beat the honeypot.
In general stings, they say they find somebody on Tinder.
uh and um and then uh and then they start revealing like the secrets to them and you see this in like some of these undercover videos and i always felt like man like if you're on a date with somebody and they immediately start telling you your secrets from asking you to tell them your secrets from work maybe you shouldn't like that's a weird date to me don't i feel so many people fall for it i don't know i like to think if you are a trained honey potter you're probably pretty good and savvy at how to extract that information.
You're not just saying, tell me your secrets from work.
You'd start with, how was your day?
Oh, what was wrong?
What did your boss do?
I think.
You're, I mean, I'm unsettlingly good at this, Ranjan.
I know.
So I think let's start.
This is making me uncomfortable.
This is making me uncomfortable too.
But that is okay.
That's how it can work.
All right.
Briefly, let's talk about Home View.
So Apple has this new device.
It's called the Home View.
It's basically a tabletop Siri, kind of like the Echo Show or the Google Nest Hub.
I think that was what it was called.
They haven't made a new one since 2021.
$299.
It seems like it's a good place for Apple to play, right?
Get its voice assistant on people's tabletops.
And then...
Over time, as the capabilities increase, you could just update the screen and the thing does more for you and you don't lose that space to Amazon's or Google's.
What do you think?
I actually, I have both HomePods and Alexa's in my apartment.
I just wasn't, this wasn't interesting to me.
I'm going to be honest, like I don't get it for Ternus's first big release.
Just let the duo hype come in and just run with it.
But like, it's just so like, You can't operate apps.
It apparently operates like CarPlay effectively.
I think there's like some number of, a very small number of apps that would be kind of native to the platform.
But otherwise, you can hand off your iPhone.
Good.
You can FaceTime.
That's nice.
Even the thing that killed me was in one of the demos they did, it's for cooking.
But you have to keep touching the device to actually like switch around and scroll a carousel.
The whole point of...
Cooking AI is to be able to just speak to it.
So I don't know.
This was like very underwhelming.
And again, at the core of it, if Siri doesn't work well, this could be so much more valuable if Siri worked.
And it still doesn't really.
Yeah, the Wall Street Journal actually did this story about it.
They said, basically, we've tested.
Siri, Siri's AI still stumbles over basic requests such as playing songs and placing calls according to Wall Street Journal tests, which I think was just Rolf Winkler yelling at his phone.
But yeah, it's got to get better at that.
But I think it will get better at that.
Remember, standard models are making everything better.
I don't know.
The market seemed to like it.
Apple is above $5 trillion now for the first time.
I know.
I mean, the amount of money they charge me in services that I don't even know what I'm spending on.
I get the valuation, but this was, I'm waiting for the duo.
I'm interested.
I don't know.
It's in eight days.
I might pull the trigger, but, and again, which is ridiculous that for as much shit as I talk on Apple, I'm about to give them $2,300 or whatever it is for that phone, but HomeView, not interested.
Okay, before we go with the debate I've been waiting to have with you for a while, let me introduce it.
The New York Times says why AI food slop looks so gross.
And then the story art is awfully disgusting.
Let's see, like artificial cream with dots all over it.
All right, but anyway, let me just read the story and I'll give you a chance to respond.
Fast food businesses in London, New York, and elsewhere have increasingly turned to generative AI to promote their dishes.
These image often...
These images often have an uncanny, unrealistic quality, and it is turning many hungry customers off.
People have been flooding social media with examples.
Pastas covered in inexplicable black dots, shapes and forms that are supposed to add up in a sandwich.
A grayish burrito that would trigger anybody's tripophobia.
The fear of looking at clusters of repeating patterns of small holes, conjuring the demogorgon in Stranger Things.
Ranjan, I think it's time for you to admit that this is a plague on our society and we need to have some, if we're going to have any AI regulation, I think it should begin with the AI food science.
Bandom, what do you think?
Vehemently disagree.
To me, around me, I remember it was right at the corner of where I live in here in New York City.
There's a halal cart.
It's probably a year ago.
I see these absurd AI images.
But he actually named his burgers and named the different chicken and lamb dishes.
I forget what they were, but there were these kind of absurd names, like the Texan cowboy something, big AI, very clearly AI-generated images.
And it was beautiful to me.
It was beautiful.
This was like someone working at the halal cart that...
If they ever imagine them having to go find a graphic designer in the past and imagine them some, and I say this, like I've worked with a lot of graphic designers, especially when I ran a startup, freelancers, like them going back, charging them a lot of money.
It was something they were never able to achieve before.
And now with the click of a button, with just a prompt, they can actually be creative and unleash creativity in ways that they never could have done before.
And you know what?
It might look a little sloppy.
Most of these are probably done on older image models where it's gotten a lot better, and it will get better, but now they are empowered that no longer are they limited to not be able to talk to a graphic designer.
Now they themselves can actually be empowered to print food photos.
So I think that's beautiful.
I don't know.
Maybe you hate America.
Maybe you hate America and the American dream.
Well, listen, I hadn't heard the Ranjan Roy explanation here.
I hadn't gotten the Ranjan guarantee on AI food science.
And now I might be swayed.
Honestly, that was a beautiful speech, Ranjan.
I love it.
I am now almost on board with your viewpoint here.
And the one thing I would say is what we should see from the AI companies is a prompt.
And the prompt would say, hey, Halal cart owner, it looks like you're printing a food sign.
Since our AI model generator makes every food sign look the same, you know, maybe you can hit a randomizer button and it can look a little different than the rest.
And then I think we would get to a much better place because all these chat CPT signs look exactly the same.
I need a little variation.
I need some character.
Do you know what?
Like, AI industry has a PR problem.
You just set them up, Alex.
That is the easiest thing to kind of build a tool in there.
Like someone is saying, I need to, I mean, let's just, let's be real, like chicken and rice.
Anytime that prompt comes in, send it to the randomizer and try to add a little variation.
And again, especially if they're saying this needs to be print quality, help guide a little bit.
People will love you.
People will love you forever.
You're going to be empowering and enabling the American dream.
Add in the tool call, guys.
Yes.
And I want to end with this because, you know, talking about how the prompt matters, this is the most beautiful example of that.
So somebody at a spa must have prompted the model, make a sign for my spa that includes things like full body massage, neck and shoulder massage, foot massage, et cetera.
The model, make an image of a sign.
The model.
must have taken that super literally because instead of making the image of the sign it made an image of a sign outside a spa so it's this picture of a sign in front of a spa that has all the treatments but in this sign image is the actual spa and it's a picture of a sign standing up on a sidewalk this person said good enough They printed it out and they put it in another sign, in an actual sign.
So what you see as you walk past this spa on the sidewalk is a sign on the sidewalk with a sign of a sign in front of a spa.
Did you actually see this?
No, I found this on, I didn't see this in real life.
I would have been, I wish I would have been so lucky to see this in real life.
I saw it on X.
This is beautiful.
This is, again, I love this.
I love, this is so ridiculous.
I love this.
It's attention.
What is the purpose of kind of sidewalk signage to grab your attention?
And do you know what this just did?
This grabbed people's attention like nothing else.
This would, there'd be some creative agency that would charge you like.
250K and there'd be a grand Don Draper style pitch and they'd be like, listen, the ultimate sidewalk signage, a sign of a sign in a sign.
And everyone would just be like, holy shit, you're amazing.
Yeah.
This stuff is beautiful.
Well, Ranjan, this was one of my strongest held AI opinions and you've totally reversed it.
So I'm on board.
Keep the AI food signage coming.
Just use a newer model, everyone.
That's all.
Yeah.
Get the prompt right or hit, you know, just put in the context window, make it a little bit more random.
All right.
Well, that's a fun way to end it, Ranjan.
That's a fun way.
I might go get a hit up the whole all cart for lunch today.
I might go.
If I ever see a spa with a sign of a sign in front, I'm going right in for a massage.
I'm right in.
And it's like, oh, man.
I just walked into another honeypot.
Always get you with the sign of the sign.
Always get you.
That's actually the ultimate honeypot technique.
All right, Ronjan, have a good weekend.
Thanks so much.
Have a good week.
All right, everybody.
Thank you so much for listening and watching.
Stay tuned for Wednesday.
We have a special interview coming, most likely.
So very excited to share it with you.
And we'll have to keep you in suspense as to who it might be.
All right.
Thanks for listening and watching.
And we'll see you next time.
on Big Technology Podcast.