NASA Gets Phished by Chinese - 2026-04-27

Summary

Talkin' Bout [Infosec] News: "NASA Gets Phished by Chinese"

Black Hills Information Security • April 28, 2026

---

Episode Overview

The panel from Black Hills Information Security (BHIS) gathers for their signature banter and analysis of cybersecurity news. This episode covers a smorgasbord of infosec stories ranging from high-profile arrests (Scattered Spider), border surveillance with Mexican vendor Seguratech, crypto-phishing in the Strait of Hormuz, and a detailed breakdown of current supply chain attacks. The conversation, as always, is peppered with running commentary on the ever-present AI arms race, including the juicy Mythos breach story and the question: Who will "win" the AI economy?

The tone is irreverent and sharp, mixing deep technical knowledge with comic relief and skepticism about infosec marketing and the current state of the tech industry.

---

Key Segments and Insights

1. Tech Giants, Monetization, & the AI Platform Wars (00:01 - 02:43)

• Opening discussion pokes fun at OpenAI’s business losses, with panelists joking that selling a hot dog can make you richer than OpenAI.
  • “If you made $1 from selling a hot dog last year, you made more money than Open AI.” – A (00:10)
• Talks about tech companies burning cash for years before achieving dominance (Amazon, Uber), and the idea that the real winners in AI might be the infrastructure and platform providers.
  • “Instead of being the AI, they’re just the platform that AI is on.” – A (00:46)
  • “There’s almost nothing special–no patentable special sauce... everybody eventually can just run their own local models. So how do you get a monopoly in this space?” – B (01:17)
• Reflection on the lack of truly useful "distilled" (open-source-clone) models and the shifting economics once major players hike prices.
  • “When Anthropic raises the rates... you will start paying more attention about those other models.” – E (02:36)

---

2. The Scattered Spider Arrest & Lessons in OPSEC (07:14 - 12:28)

• News punchline: Tyler B., member of the infamous Scattered Spider group, pled guilty and was extradited to the US. The panel explores his bizarre OPSEC failures.
  • “He logged into Namecheap from his home IP.” – A (10:37)
  • “You deserve it at that point if you’re this prolific... and your OPSEC is that bad.” – C (10:43)
• Panelists highlight the trend: successful cybercrime groups often have operational slip-ups that lead to their undoing.
  • “It’s almost Infosec Jackass. What’s the dumbest thing you could do?” – B (11:35)
• Comic relief about cybercrime geography, extradition risks, and comparisons to “hacking from Kansas.”
• Reflection: Big companies are still worryingly easy to breach; most of these threat actors are very young and reckless.

---

3. Cross-Border Surveillance and ‘Intelligence Laundering’ (13:02 - 18:18)

• Key Story: The U.S. contracts with Mexican surveillance company Seguratech, sparking concerns about data custodianship and privacy.
  • “Countries hire companies from other countries to get that data instead... It’s like money laundering, but intelligence laundering.” – B (15:04)
  • “Mexico is spending about 5% of the entire global spend on surveillance tech... three times what their population would suggest.” – D (13:49)
• Panelists mock the dystopian “control room” visuals and share their suspicions about surveillance circumventions.
• Sidebar into comparisons with “The House of the Scorpion” novel, emphasizing recurring themes of outsourced power and accountability.

---

4. Strait of Hormuz Phishing: A New Level of Themed Scams (20:25 - 22:56)

• Current Event: Scammers text vessels promising safe passage through the Strait of Hormuz—for cryptocurrency.
  • “It makes more sense... they’re phishing with people on boats!” – E (20:40)
• Jokes abound about the boat-themed phishing and how the chaos of real-world conflict provides new opportunities for cybercriminals.
  • “Send us one Bitcoin and we’ll get you through.” – C (21:37)
• Reflection: The scam blends just enough plausibility and chaos that some companies have fallen for similar schemes previously.

---

5. Supply Chain Attacks: Bitwarden CLI Compromise (27:48 - 28:56)

• Incident: Bitwarden’s NPM package (used for their CLI) was briefly compromised in a supply chain attack.
  • “It wasn’t wild for very long... but if you were using the Bitwarden CLI, check if you pulled the compromised version.” – E (28:04)
• Small % of users impacted, but panelists point out that CLI attacks typically target more technical users, increasing the risk of secrets exposure programmatically.

---

6. Culture & Nostalgia: Cult of the Dead Cow’s ‘Reboot’ (30:23 - 33:39)

• The panel lampoons the relaunch of legendary hacktivist group Cult of the Dead Cow, expressing skepticism about its relevance today.
  • “If they called themselves anything else than restarting Cult of the Dead Cow, no one would give a shit.” – B (32:15)
  • “What I want to know—what was the final straw? Was it AI-generated Spiderman videos? Having to tip 20% at coffee shops?” – A (32:51)
• Underlines the tension between infosec nostalgia and industry progress.

---

7. NASA Gets Phished by Persistent Chinese Threat Actor (34:13 - 36:18)

• Headline Story: Long-running phishing campaign, attributed to a Chinese APT associated with the aerospace industry, successfully phished NASA personnel.
  • “Targeting professors, researchers, and engineers... asking them for plans, pretending to be colleagues. And boom—got some secret stuff.” – F (34:58)
• Strong social engineering and careful targeting (customizing emails based on friend groups and personal relationships) were key to success.
  • “There was very, very good OSINT done... masquerading as people’s friends.” – F (36:07)
• Parallel drawn to infamous leaks on gaming forums (e.g., “War Thunder” reference – 35:55)

---

8. Microsoft Teams Phishing and Modern Social Engineering (37:24 - 40:59)

• Mandiant Uncovers "UNK6692": A new social engineering campaign heavily abuses Microsoft Teams as the phishing vector, chaining together various “living off the land” techniques.
  • “Teams is the new version of a phishing phone call... lots of threat groups are abusing it.” – A (37:25)
• Rapid chaining of multiple techniques leads panelists to suspect some AI-driven orchestration, though nothing in the campaign is technically novel.
  • “There’s nothing really super cutting edge... It’s just the speed and the chaining together that leads me to believe they might be using some AI.” – B (38:32)

---

9. The ‘Mythos’ Leak: Anthropic’s Secret Model and the API Key Fiasco (41:19 - 46:33)

• Big Story: Threat actors briefly accessed Anthropic’s “Mythos” model via compromised 3rd party vendor API keys.
  • “People claim to access it... but they only ran queries through exposed API keys. Anthropic knows exactly what they did. It’s more stunt hacking than actual impact.” – A (42:54)
• Rumors swirl about how the breach happened; the real risk was hyped, but actual model weights weren’t obtained.
• Panelists compare “Mythos” to OpenAI’s release behavior.
  • “When you call something Mythos...that is a much stronger mnemonic device... It’s about marketing.” – B (45:23)
• OpenAI’s “GPT-5.4 Cyber” and “GPT-5.5” quickly come up as competitive, reflecting the current trend of “benchmark wars.”

---

10. The Global AI Arms Race & Defender Dilemma (46:33 - 49:54)

• Indian Government Advisory: Warns that AI is accelerating exploit chains; vulnerabilities are now weaponized in hours, not weeks.
  • “It pretty much just says... vulnerabilities can be chained by AI incredibly easily... within 24 hours is now the benchmark [for patching].” – A (47:44)
• Conversation shifts to how next-gen AI models (like GPT-5.5 and Claude with longer context windows and auto modes) will further supercharge autonomous attack and defense cycles.

---

Overarching Discussion: Who Wins the AI Arms Race? (55:13 - 66:15)

Roundtable Debate: Which entities will "win" in AI—platforms, open source, hardware makers?

• Infrastructure over Models: General consensus that “winning” in AI will look more like winning in infrastructure and platforms (à la Amazon, Nvidia, ASML) than in service providers (OpenAI, Anthropic).
  • “If they start raising their prices... We have the skills. I’m going to spend a quarter million, build my own infrastructure, and we’ll just build it out. I don’t see how they win.” – B (56:35)
• User Perspective: For regular folks, the best approach is to avoid vendor lock-in and maintain flexibility across models and services.
  • “You have to be decoupled from these ecosystems... have things and skills that are transferable.” – C (61:54)
• Economic Model Prediction: Future AI access will mirror streaming/content services: premium, ad-based, and open-source or self-hosted options, with Google predicted to monetize AI most efficiently.
  • “It’s going to be the same as every other ecosystem... The more ads you’re willing to stomach, the less you pay.” – A (64:15)
• Open Source Progress: New research (e.g., TurboQuant) will increasingly make local, affordable, high-performing models feasible for end-users.

---

Notable Quotes & Moments

On Cybercrime OPSEC

• “You logged into Namecheap from home. Oh, geez. That’s gonna hurt.” – B (11:35)

On Cross-Border Surveillance

• “It’s like money laundering, but intelligence laundering.” – B (15:04)

On AI Value Proposition

• “I can have it analyze birds and burn a thousand dollars worth of usage for 10 months.” – A (50:31)

On the Future of AI Providers

• “All of these companies... are thinking, 'We’ll be the Google of AI.' That’s where they want to be. The problem is, none of it is patentable.” – B (55:13)

On Personal AI Strategy

• “The better project manager you are, the better your output will be [with AI]. It’s all about defined requirements.” – C (69:26)

---

Humorous & Memorable Exchanges

AI Burn Rate

• “How much money have we spent this month on AI?” – B
  “If it’s not big enough, that means you’re not using enough of it, which means you’re not going to succeed.” – E (50:18)

On Vendor Nostalgia

• “If they called themselves anything else ... other than Cult of the Dead Cow, no one would give a shit.” – B (32:15)

Sci-Fi Dystopia

• “The winner is the one that destroys humanity.” – B (59:39)

---

Timestamps Guide

• 00:01 – Opening banter; AI startup economics
• 07:14 – Scattered Spider arrest and OPSEC mistakes
• 13:02 – Seguratech Mexican border surveillance
• 20:25 – Strait of Hormuz phishing scam
• 24:37 – Signal/iPhone notification bug and its fix
• 27:48 – Bitwarden CLI supply chain attack
• 30:23 – Cult of the Dead Cow reboot
• 34:13 – NASA Chinese phishing campaign
• 37:24 – Microsoft Teams as a social engineering attack vector
• 41:19 – Anthropic’s Mythos model leaked (API access, not weights)
• 46:33 – Indian government: AI speeds up vulnerability exploitation
• 55:13 – Who wins in the AI arms race? Debate
• 61:31 – How the “common person” survives the AI arms race
• 66:15 – Local vs. cloud-based AI models
• 69:26 – Human-in-the-loop project management and quality issues

---

Episode Takeaways

• Cybercrime still comes down to OPSEC mistakes, even with state-of-the-art tooling.
• Nation-states and third-party surveillance create compliance "gray markets" for data.
• Phishing, in all forms, is ever-adaptive—now including timely, region-specific scams.
• AI is both a tool for good and a force-multiplier for attackers; arms race logic dominates.
• The winners in AI may not be the model creators, but the platform, hardware, and infrastructure providers.
• For users and companies: stay agile, avoid lock-in, and be prepared to pivot strategies as the landscape shifts.

---

Stay tuned next week for more infosec news, wild speculation, and generative AI hot takes from BHIS and their rotating ensemble!

Transcript

[A]

Dude, are you telling me I make more money than open AI?

[B]

Yes, you make more money.

[C]

I love that logic.

[D]

I make more money than open AI.

[A]

If you. If you made $1 from selling a hot dog last year, you made more money than open AI. Anyway.

[E]

Well, so it's not uncommon. Not not. It's not uncommon for these big businesses

[A]

to burn a ton of cash. Yeah, yeah, I know how economics work. Blah, blah, blah. We had Amazon not turning a profit for 10 years, and now Bezos is buying 15 million and now you want

[E]

to go look profit sheets per quarter. It's like, disgusting.

[C]

What the numbers are. Right. Like, it is.

[B]

It is Amazon. They're like, you all fight on the AI thing, and when you get it sorted out on our infrastructure.

[A]

Yeah. Instead of being the AI, they're just the platform that AI is on.

[E]

I have one good example, like Uber.

[C]

Right.

[E]

They were burning tons and tons of capital, and I think they still are. But the points I'm trying to bring up here, they had to raise their prices because everything was so cheap. It was all like, great in the. The beginning, but they got to make money eventually. And that goes the same for AI, Right.

[C]

Yeah.

[A]

Well, also the monopoly, that's the tech. That's the tech play. You push down all the prices, then you monopolize it and you push them back up.

[C]

Yeah.

[B]

None of these bastards are going to do that. Right. Because like, literally there's nothing. I hate to say this, but there's almost nothing special. Like, there's no, like, patentable special sauce.

[E]

Yeah.

[B]

That they're going to get. So. And everybody eventually can just run their own local models. They can run it on their own infrastructure if they want. So how do you completely get a monopoly in this particular space in a way that, like, Anthropic or OpenAI wins the AI? You're not going to.

[C]

And that's why they kept Mythos secret for a while. A lot of people hypothesize is because their Chinese model would just distill it and then release it open source within a couple years. Right. So if you keep a public model, it'll get distilled.

[A]

That's true. Okay, hold on. Name a distilled model that's actually good.

[C]

I mean, composers distilled off of like, anthropic, isn't it?

[E]

No, no.

[A]

Composer was.

[C]

Composer was supposed to be.

[A]

It's actually.

[C]

Found it in the source code.

[A]

It was Kimmy. Oh, that's right.

[E]

I saw that.

[A]

That's what I'm saying. Okay, so my question is name a Distilled model that anyone would actually use.

[C]

There was one. I can't remember the name.

[E]

I'll use a Chinese model if it's cheap.

[F]

I was about to say no one's using it.

[B]

It's from Canada. You would know it.

[A]

Quinn isn't distilled, is it?

[F]

No, I don't know. I just. I just.

[C]

Literally just.

[A]

But here's the deal, though.

[E]

Here's the deal, Cory.

[A]

If.

[E]

If anthropic. Not if. When anthropic raises the rates enough, you will start paying more attention about those other models.

[A]

Yes, you're right. Also, we did start this conversation by saying, how can we not talk about today on the news?

[C]

Apparently, the Mini Max models were just.

[B]

I'm liking about to be arrested. Like, there should be sirens behind me.

[A]

Yeah, I think you're in a shipping container. Hostage. Don is being held hostage.

[C]

It's like that clip where it's like the dude from Harry Potter that got arrested and in his mugshot he's, like, freaking out. Like, that doesn't help when you look innocent.

[A]

Guy, send us a picture. I want to see this. I don't know what you're talking about.

[C]

Of the Harry Potter thing.

[E]

Okay.

[A]

A guy from Harry Potter got arrested. Is it Snape? I knew he was the bad guy the whole time.

[C]

It's. Oh, here we go.

[F]

It was. It was one. It was one of serious Black is

[B]

legit because that would be less controversial than Harry Potter right now.

[C]

Oh, my Discord needs to do 42 updates. One moment.

[E]

Yeah, that's normal. Actually.

[A]

Only 42. Dude, it's just distilling your model.

[C]

Yeah, yeah, dude, I wish it would distill me right now.

[F]

Wasn't it one of Draco Malfoy's, like, henchmen? It was one of those two guys.

[C]

It was. It was Harry's godfather, the one that got arrested and was, like, tripping out in his mugshot.

[B]

Gary Oldman.

[F]

It's Gary Oldman. Yeah.

[C]

Oh, my gosh. Okay, now it's doing eight more updates.

[B]

AI is getting so good at putting my face on random things. I'm really learning who I actually look like.

[F]

Like, you could pull a RoboCop. John. John RoboCop themed human Android or, like,

[B]

you know, like cyborg. I. I just. Yeah, that's great.

[F]

Yeah.

[D]

There's the picture.

[A]

Yes, I see the mug shot. That looks like it's from the actual book, though. I mean, the movie.

[C]

It might be. It's. It's from. It's from Google Images, apparently. Pinterest. So that's why it looks like garbage. Quality.

[D]

As far as I know, that from the movie. Didn't have an image for that. That is from the movie.

[C]

Well, it is from the movie.

[E]

Yeah.

[F]

He's screaming because the dementors are eating him.

[A]

Like, I was about to be like, dang, this guy got a dui. What is going on? All right, should we start the show, or should we just talk about AI for 20 minutes?

[B]

Let's start the show. No, he's driving. Go to him.

[A]

So go for it. Roll the finger. Do it. Hello, and welcome to Black Hills Information securities. Talking about A.I. wait, no. Sorry, wrong show. Talk about news. It's April 27, 2026, and we're here to talk about all kinds of fun things. We have scattered spider people going to jail, pleading guilty. We've got Mexican surveillance giants taking over border control. We've got phishing text based on. Are you trying to pass the Strait of Hormuz? Click here. We've also got access to Glasswing, AKA Mythos. No, we don't, but some discord group claims that they did, so we'll get into that.

[C]

I didn't see the straight of Hormuz 1. That just seems like an AI was told, like, hey, go fish, people. Based on current events. And it was like, you got it, boss. I'm on it.

[A]

I'll run a quick round of introductions. There's only six of us today, so our Brady Bunch vibes are key. Hayden to my. Wait, Hayden, who's our sock person? We've got Aisling, who is our resident, I guess, flammability assessment coordinator.

[D]

I have lit a lot of swords on fire in my day.

[A]

John Strand, who's not tweeting about how much he's spending on Claude, but could be if he wants to get that VC money.

[E]

Oh, that's sweet. VC cash.

[C]

You could be living the dream.

[A]

Ralph, who is currently building a raspberry PI or something. I don't know. Who knows? Dallas building his own raspberry PI.

[B]

Could be like a Rubik's Cube.

[E]

AI pie.

[A]

An AI They've got Wade, who's ordered some pizza from the Bhis Hack and Fresh Webcast pizza company.

[F]

It's excellent. No pineapple. This is a no pineapple.

[E]

Oh, I love pineapple Pizza House.

[A]

It's personal choice.

[B]

An apple on pizza thing.

[F]

Okay, you said you didn't want to be controversial, John. Here we go.

[E]

Here we go.

[A]

Should we just do the AI should we spend the whole first part of the show talking about AI or the whole show? Let's do second. Let's. We'll do second half. So.

[C]

Because on the second half they'll cut us off.

[A]

We'll do this non AI articles first. If you hate AI, you can leave halfway through. That's perfect. So first of all, there was a Cribs on Security article basically talking about how Tyler B, one of the members of Scattered Spider, pled guilty. He's a UK citizen and was actually like fleeing the police and was like in Spain and then he was flying to Italy. I like how the picture they're using of him is from when he was like seriously 12.

[E]

Yeah.

[A]

Why is that?

[F]

That's when he was active.

[A]

Well, we know what he looks like.

[F]

When he was actively happy

[E]

with that.

[B]

That guy rolled out with like a guitar. I'm like, this song is about to just slam or like he looks like. He looks like. What is it the Hives or whoever. Like when like really kick ass music was back in the early 2000s. But that's a horrible picture.

[C]

He also called him a computer geek. I didn't know he still used that word.

[A]

I know this is like the. The picture there is a very much like I feel like I'm at like a Sainsbury. So something in the United Kingdom. Like reading basically the. I'm reading the equivalent of like the National Enquirer.

[F]

Like.

[A]

Right.

[C]

Yeah.

[A]

He's just got the article

[B]

and then it's off. Like it's just.

[A]

That's only.

[B]

Only four people in the audience will get that reference.

[C]

That's it.

[A]

So the, you know, the article is pretty in depth, but basically it just runs through the history of Scattered Spider. For those that don't know, they breached, you know, MGM and Marks and Spencers and a bunch of other companies. They had SIM swapping and a bunch of other members. It was an English speaking cybercrime group. So which kind of thought.

[B]

Right. With the sophistication of the social engineer, we thought that they were English speaking

[A]

from the start, so. And just to be clear, he was extradited is that. That's the other thing that we're kind of not really covering here is that he was. Yeah. Dragged to court in the US and there's the picture of him being dragged by some Spanish police to the us.

[B]

So the lesson to be learned here is don't commit crimes where you could be extradited from.

[E]

That's like a lot.

[B]

These cyber criminals are like in Kansas, you're like, what the hell were you thinking? Go to.

[A]

All of them were too like the group is. Okay. So the group is a guy from Florida. Guy from Florida. A guy from Texas. A guy from. Another guy from Texas. A guy from Jacksonville, North Carolina. Another couple from the uk. Like, I don't know.

[C]

I thought you were kidding.

[A]

No, no.

[B]

Really?

[E]

No.

[A]

Like, they were so successful.

[C]

Really?

[B]

Wow. So.

[A]

So, yeah, I mean, I guess in

[B]

this one situation, out of the billions of dollars that were stolen and all the damage, these guys were caught. So that just proves that crime doesn't pay.

[A]

Yeah, it doesn't pay.

[C]

It pays for a little while, and

[A]

then it doesn't pay. Because you can't pay for AI with Bitcoin.

[B]

No.

[E]

You should be able to, though. Why can't you do that? That's a good question.

[C]

Let's start.

[A]

Because of kyc, man. We can't have nice things.

[E]

All right.

[D]

Anyway, that's not what KYC stands for.

[A]

KYC definitely stands for no Abusing AI if we are.

[E]

Yeah, yeah, it's an acronym, but you have to look deeper.

[C]

It says they linked him through, like, reused usernames and email addresses, but.

[A]

Oh, dude, he. No, no, it gets worse. He logged into namecheap from his home ip.

[C]

Oh, my God. Okay, well, you deserve it at that point if you're, like, this prolific, like, hacker group and your OPSEC is that bad.

[E]

Well, he didn't think he was going to get caught.

[A]

These kids are 16 years old. Are you telling me they don't have fully developed frontal lobes? Turns out they don't.

[C]

I guess they're developed enough to hack these giant companies, but not to use a different email

[E]

from a. From a. What do you call it? A mind perspective here. This is about risk. Like, they don't care. They don't think about consequences.

[A]

That's the same thing as me riding a motorcycle with no helmet.

[E]

Yeah, because it's cool.

[A]

Because it's cool. But also the. The group, I think, in general, the comm. Whatever you want to call them, their basic thing is, like, this shouldn't be this easy to hack these big companies.

[E]

That's what I keep doing.

[B]

I almost feel like this is infosec, jackass. Like, what's the dumbest thing that you could do? And then you're like, don't, don't. No. Oh, God damn. You logged into namecheap from home. Oh, geez. That's. That's gonna hurt.

[C]

There's playing chicken to see who gets caught first.

[B]

Yeah, it's like, who's doing the dumbest thing?

[D]

Kind of. Yeah, it really gets.

[A]

I'm not kidding.

[D]

A lot.

[A]

Yeah, I'm not. I think that's pretty accurate.

[E]

I mean, I think it's good that they got caught because it actually shows that there is some kind of punishment that could come along.

[F]

Yeah, you don't hack from exactly what John said. You don't hack from a state that has extradition treaties. That's it.

[E]

Yeah, exactly. Maybe the whole country.

[A]

Not a real high bummer. Or if.

[D]

Or if you're gonna do it, you get the heck out right away. Like, you go, okay. It was easier because I was inside the, you know, firewall boundaries or whatever.

[A]

Bye.

[E]

Yeah, without them being arrested, then nobody else would know.

[B]

These cybercrime groups don't f with China because I think that China has a completely different way of dealing with things.

[F]

The Chinese executed a dude who is fishing.

[B]

Like, they don't extradite.

[A]

Actually fishing or fishing.

[F]

No, fishing with the p. He got caught scamming and fishing. They just started. Yeah, I got executed now.

[A]

So he was sending phishing emails while illegally fishing in this. In the South China Sea.

[C]

Didn't have a fishing license.

[E]

You got to get one of those,

[C]

work for the state.

[A]

What's next we could talk about. So this is kind of interesting. So basically, it looks like the US has contracted with a company called Secure Securitech Seguratech, which is the. A Mexican company that has pretty broad surveillance capabilities to monitor the U.S. border. I don't really know how big the contract is or what it is, but essentially everyone's freaking out because they're just assuming that this is kind of sketchy, that the company isn't properly secured or doesn't have proper data because they custodianship and all that stuff. But yeah, I don't know. It's an interesting angle to think about, like, hey, watch our border for us. And by the way, we're going to pay you a huge, you know, contract fee to do that.

[D]

It's kind of fun. The 1.27 billion number is not, like, how much they're spending this year, but a lot of it comes out of not a lot of years. That all pesos, right? It's initially an uninflated pesos. So this is just tallies over the last, like, 10 years or so. But the thing that caught my eye is apparently Mexico is spending about 5% of the entire global spend on surveillance tech, most of it going to this one company. And I went 5%. That seems like a lot. Isn't Mexico small? And I checked and, like, it's three times the fraction that their population is out of the world. Like, they're 1.6 and 1.3% of the whole world, depending whether you're counting people to have cameras on or land mass to have cameras on. But one way or another they're spending like three times how many people they've got worth of the global share of spend on this. Also look at this building. Is it me at all or does that look like the tower in the middle of a prison yard? Except this is the entire city looks like Arasaka tower.

[B]

Gary Bantam Panapticon at that point. So, so I want to talk a little bit of background on this. So what happens whenever you're dealing with intelligence, like in the United States you have something like conus, which is a continental United States and you cannot do surveillance on US Citizens without a warrant. Right? So one of the things that countries do, because a lot of countries have similar laws, is they will hire companies from other countries to get that data instead. So that's a way of circumventing it. So the United States, instead of actually directly monitoring US Citizens, they acquire data and like surveillance data and things like this from various third and third party independent companies or completely different countries to do it on our behalf because it's illegal for them, them to do it through this direct means. But if they buy it through these third party services, then it's, I guess it's kind of like money laundering, but intelligence laundering.

[A]

I think that's a good analogy.

[B]

This is really common. So you may have something where somebody, okay, says well we were monitoring data and it looked like somebody in the United States is doing something. And the United States can open up a FISA warrant at that point and be like, well, we weren't monitoring, but someone said now they can start monitoring that individual. So this didn't surprise me all that much, especially because of what's going on on the border. And now they can just acquire the surveillance directly without having to do the surveillance themselves. Follow the law.

[F]

Can we scroll down to the one image? Like if you go to the news article. Scroll down. Okay, right, stop right there. Look at those monitors. I want some of those. That's what I'm really looking at. You see how big they are?

[C]

That's just a TV.

[A]

You can't call that a monitor.

[B]

Is it a TV?

[F]

Is that a TV like a 4K

[A]

TV in your lap? It's a 43 inch TV. No one's stopping you.

[F]

It looks like it's a map on that thing, dude.

[C]

Gets a little lasers going back.

[D]

It really looks like it's got to Be touch screen.

[F]

Right, Right.

[A]

Yeah.

[F]

Could you IM like, you're. You're pulling up logs. Enhance and like.

[A]

Oh, okay.

[F]

Then unenhance and throw the logs over there. Bring in some new logs.

[E]

Dude.

[C]

Put a pew, pew map on it, man. You have to, if you can.

[E]

If you enhance the logs enough on that screen, you can actually see the threat actor.

[F]

My cloud code would look so great.

[A]

Yeah.

[D]

Oh, yeah, right there. Ask.

[C]

Can you imagine vibe coding on that screen right now?

[A]

Dude, I cannot imagine watching my Claude usage slowly chip away on that. Oh, man.

[B]

Is this a scattered spider? I don't understand. I have read the book Scorpion King because this is starting to feel a lot more like Scorpion King. If you haven't read the book, you should. It's a fantastic book.

[A]

But have not read the book.

[B]

Gotta look, gotta read movie that's not related.

[F]

Dwayne the rock.

[D]

The rock actively making and.

[B]

No, that's not it.

[E]

I'm sorry.

[B]

Scorpion King is. There's a drug lord that's. Controls the border on the Mexican side that controls and stops immigration. But it's very, very similar to this type of scenario.

[A]

All right.

[C]

I think I found the wrong Scorpion King. This is

[A]

Scorpions.

[C]

No, I searched it on a book app and it looks like a series of spicy novels, so I'm not going

[A]

to add that one to my.

[D]

I mean, that doesn't surprise me either.

[A]

I'm gonna.

[C]

I don't think that's the one John was talking.

[A]

The amount of AI Slop.

[C]

Oh, dude, I know romance.

[A]

I can only imagine steam is just as bad.

[E]

AI Slop. There's this real writer slop in that.

[A]

That's true. I've read Moby. Writers are still a thing.

[B]

I got the. I got the book name wrong. It's the House of the Scorpion.

[A]

Oh, okay.

[E]

That's why I just can't find it.

[A]

He got the other one.

[C]

Was the other one novel?

[D]

Yeah.

[A]

John is reading another spicy romance novel.

[B]

House of the Scorpion is the name of the book. Sorry, my bad. I sent you all the spicy novels, which now I'm hoping that there's a spike of Scorpion King novels. Novel took off the way it did.

[A]

Romantic. It's like Google took me like a scorpion and poisoned me.

[D]

Look, there are a lot.

[B]

There's a lot of dinosaur like it is.

[A]

Oh, yeah. Dinosaur porn is a whole separate. That's a different podcast, John.

[E]

We are really going down a rabbit hole.

[C]

Guys, we should talk about. I think so.

[A]

All right, let's keep going tangentially. Real quick. Tangentially related to that actually was an interesting article in the New York Times. I didn't put it in the show notes, but basically this morning there was an article about how the US like, gold is tech is basically like laundered money. Essentially. Like a lot of the gold that we issue in, like US backed gold currency is actually mined in like slave labor mines and sketchy scenarios. So this is just the intelligence equivalent to that. It's like, basically there's a document that says this is mined. And they're like, all right, we didn't look any further than that. Anyway, anyway, let's talk about the Strait of Horror moves. So current events, obviously. Reuters reported an interesting article about, basically people are sending fraudulent messages promising safe passage to the Strait of Hormuz in exchange for cryptocurrency. Basically. I guess we don't know necessarily, we're assuming this is a scam, but maybe this like on a. I feel like this is fishing. Like, we've gotten to like the next level of fishing where like, it's kind of looks like a scam, but also it could be legit. Like there could just be some pirates that have some crypto and are like, hey, pay for us and we'll defend you. I don't know if that's the thing.

[E]

The irony is that they're fishing with people who are on boats. So it's like, it makes more sense. Right.

[D]

But they are, they are fishing the Strait of Hormuzzi.

[A]

Yeah, they're fishing the Strait of Hormones the other way.

[E]

Yeah.

[B]

Wasn't there like five ships that ran the blockade yesterday? I wonder if that one of them

[C]

was like a cruise ship and they're

[B]

like, oh, you're cruising.

[A]

Yeah, it was a booze cruise. Everyone was drunk as hell.

[D]

There were definitely two ships and one of them definitely got shot at by

[A]

Iranians who were like, no, you didn't pay for anything.

[B]

Set a course for.

[A]

No, no, listen, there was no warranty. Okay, okay.

[C]

So this, this basically just says that someone is, you know, is fishing or whatever you want to call it, pretending to be like Iranian authorities. And I, I thought this was way funnier when it was just like I was imagining like a grandmother being texted like, hey, do you need to get through the straight of horror movies?

[A]

I got you.

[C]

Don't worry. Send us one bitcoin and we'll get you through.

[B]

But if we get back to it, I mean, we were joking about it at the beginning. It's like Iran is charging like what, 2 million in Bitcoin?

[A]

Yeah.

[B]

Or in cryptocurrency.

[A]

That's the scam. Yeah, they're both scams.

[B]

Either way, what the real situation is

[A]

as well, that's basically the campaign is taking advantage of the chaos and basically being like, would Iran actually charge this much in bitcoin to get across? I mean, maybe.

[B]

Yeah. Not outside of the realm.

[A]

Turns out it was a scam. Who would have thought? Just like that Nigerian prince that had all that money and he was trying to give it away.

[F]

There was. There was some Middle Eastern country that you. You used to have to pay them to sell you, to sell product to them. Like, you'd have to send them a actual direct deposit. And it was a real corporate standing that they used to do in order to make sure that the sales people weren't taking their time and that they would actually set stuff up. And the reason I know this is because we got. I. I have triaged a phishing email that was based on that. And then the actual company who did that has a thing on their email address. It's like, we no longer do this. Do not send any money to anyone. Because people kept sending money to these fishers and abusing it. Yeah.

[A]

Which is wild.

[F]

Why would you pay anyone to sell them something? Right. It's like, well, that's a cultural thing.

[E]

Maybe.

[B]

Wait a minute. So with Amazon Occult, with Amazon, they. They owed us money for some stuff that we did and they had this whole entire thing. They're like, so if you want to get paid early, then how about you pay us? They have like a separate company, 5% of the total amount that Amazon owes you in order for us to pay you early. And we're like, what?

[E]

That.

[B]

Nope, that totally was the thing. And it was legit. We called her point of contact and they're like, yeah, that's another company we spun off. So what they do is literally look at who they owe money to. And they're like, yeah, if you pay us 5% of the total amount of money we owe you, we will pay you early. The funny thing was they were six months late when we got that email. And we're like, no. And then they responded back, well, if you want to get paid how much? You pay us five. It was ridiculous. So, yeah, these types of things. The reason why the scams work is because there's a bunch of companies that operate like that.

[F]

I had a. I had a corporate contract that they tried to bake that in, that if they paid me on time, they got the. A 5% discount. And I was like, no, take this, take this. Out of the contract, I'm just going to charge you 5% more. How's that?

[A]

I'll charge you 5%.

[C]

Exactly.

[F]

Right, right.

[C]

You're gonna be 5%. You know, non. Added fee.

[F]

Shout out. Shout out to chat GPT for reading that contract for me and pointing that out, you know.

[E]

Yeah.

[C]

What a great lawy, honestly.

[A]

All right, so continuing on our non AI articles, we talked a couple weeks ago, maybe last week, about essentially, the FBI discovered a way to mine the notifications on iPhones. Essentially.

[C]

That was a cool one.

[A]

There was a bug that let the FBI recover deleted signal messages because the context, the content of the messages was stored in the notifications database of the IO of the iPhone. But that has now been fixed. And so Apple published that. They fixed that, and it's now not a problem anymore. And so that's kind of fun because now I can freaking turn back on message notifications because it was awful for a week because I was just. It would always just say, signal, new message. And I'd be like, what is it? Is it a meme? Is it a fish?

[E]

Is it important?

[C]

I was going to say, could you not just, like, have your notifications off and not show up on, like, the home screen? I guess that would be so unclear.

[A]

So a couple of things.

[D]

A couple of things were happening. The messages were coming through, and for people who had set it so that you could see the message content ever.

[C]

Yeah.

[D]

Then the notifications database had what the message content was, which is basically everything.

[A]

That was the default behavior. No one was probably going in there and switching that. Very few people. Less than 1%.

[F]

Yeah.

[D]

But the real bug was not so much that this was in the notifications database, and therefore the FBI could dig it out of the notifications database. It was the fact that if you deleted an app, it didn't clear notifications that were. That were from some old app.

[A]

So they didn't just delete, they deleted the whole app.

[C]

Right.

[D]

They delete the entire app. And then the notification database stuff hadn't been touched.

[A]

That was the bug.

[D]

That's what happened.

[B]

Residuals.

[C]

Okay.

[D]

And because people, like, even if you had the. You're not allowed to see what this is on the home screen. You're only allowed to see what this is if you're actually unlocked. Not the home screen lock screen. Then it would still have the text. It would actually go into that. Like when people went, oh, well, I'll switch it up. There would just be a little notifications database saying there was notification, it was signaling.

[C]

I skipped the Middleman. Which is still memorized my notifications because they're funny.

[A]

Dude, the Apple summaries are so. The Apple AI is such a joke. It's like worse than any other AI.

[C]

It's honestly always give me a laugh.

[A]

Some wild worth turning on because of how bad it is. Oh, I know. Every suggested response it'll give you will be like, yes or no. Like, that's the only response.

[C]

Oh, thank you.

[A]

Like, I'm at Costco, what do you need? And it's like, yes, no. How is that helpful?

[E]

Summarize.

[B]

It's not like emails from me approve or deny.

[A]

That's all it says on John's email.

[B]

It's approve or ignore buttons. Approve, deny. That's it.

[A]

I like Hayden's version. Approve or ignore. Yeah, I heard.

[E]

I heard Apple's gonna get spicy this year. They're gonna partner with X for their AI.

[A]

That's not true. They already partnered with Google. Don't be dropping fake news.

[E]

All right, come on, man. Let me first.

[C]

You got stuck in Twitter, man. What are you trying to do?

[F]

I know, right? Right. Look at.

[A]

Get out of here.

[F]

Oh, no.

[D]

Yes. No is a binary set and can be used to encode strings.

[A]

So let's talk about the Bit Warden CLI thing. I think that was pretty hot. Ralph, you want to dig us through that one? I feel like you're, you're, you're, I know you're like a Bit Warden user, so this probably hit you pretty hard.

[E]

Actually, actually, I'm not a Bit Warden user right now, but there's nothing wrong with Bit Warden. I, I, I'm using 1Password right now, but there's nothing wrong with Dead Warden.

[C]

But anyways, I prefer one password. Yeah.

[E]

Yeah. Their NPM package got compromised and their NPM package is what they used for their cli. So the bitwarden cli. And it was another supply chain attack, which we've already done, like three or one checkmarks.

[A]

Yeah, check marks. Which checks marks is probably a serial. Sorry about that.

[E]

Yeah. So if you were using the Bit Warden cli, you should definitely, you know, see if you had updated the NPM package recently or got this compromised version. I don't think from what I saw, it was actually in the wild very long. They did detect it pretty quickly.

[C]

It was like a few hours, I think.

[F]

Yeah, yeah, they did.

[E]

Yeah, it wasn't, it wasn't wild. But there's a very specific use case, so it's not someone who is just using it. You'd have to have used the CLI Yeah, yeah.

[A]

The CLI is probably a small percentage of the actual.

[F]

I'm gonna tell you, if someone's using the cli, it's most likely programmatically. There's more secrets. That's why.

[A]

Oh, yeah, you got to have Dune references also.

[F]

What.

[A]

What is.

[C]

Adam. Maybe I don't give them the attention. This company, whenever they put out an article, I see it on osint. They started advertising like they're reporting on these sorts of things, and that just is so wild to me. You put out, like, a CTI article and then make it an ad, like, literally on Twitter.

[A]

Standard for all security.

[C]

Well, no, it's like an ad promoted on Twitter kind of ad.

[E]

Oh, to get you to go read the article on there.

[C]

Yeah, yeah, yeah.

[E]

Well, they want to drive traffic. I mean, I'm sure they're selling something else there. Says pricing at the top, Right?

[A]

So that's a pretty good way for. To make sure that I can't see your ads because I block all the ad.

[F]

Or.

[A]

Sorry, see your articles because I block all the.

[C]

I'm gonna put pricing at the top of my personal website and see if anybody clicks it.

[A]

Wait, how do I buy Hayden? What's the pricing on that?

[C]

It depends on how my week's going.

[A]

So it's dynamic pricing.

[F]

I wouldn't want to pay for Hayden's Claude. Claude account.

[D]

Right?

[F]

Like, if you buy Hayden, you have to also pay for Cloud.

[B]

I know the numbers, so it's.

[C]

Listen, I only hit my quota a few times this week so far. All right?

[A]

Yeah, dude, it is hard this week so far.

[E]

You have a call.

[A]

So far it's Tuesday, so Monday. Continuing. This is kind of an article, a non article. I don't know. I don't know if anyone really knows or cares about this, but apparently the Cult of the Dead Cow has rebooted themselves as of April 21st. I don't know that much about them or care. Does anyone have an opinion on this? John? You might.

[B]

I don't know who.

[F]

Who.

[B]

Who's in it. Like, who's. Who's in the article? Because one of the problems I have is, like, there's.

[A]

There's kind of Oxford roughing, dude.

[C]

Oh, yeah.

[E]

Oh, yeah. No, I totally don't know. Now.

[A]

I played color with that guy called Dead Cow dot com.

[B]

But if we scroll down, like, who are the members? Right? Because, you know, there's all kinds of people that were associated with Cult of the Dead Cow and the previous Loft Heavy Industries Group and things like that. But the problem I have is like, you can't go to a conference without running into four or five graybeards. Like myself, they're like, I was part of Cult of the Dead Cow. And it's like, no, dude, you were just dead DEFCON when back Orpheus 2000 was released. That doesn't mean, you know, close, but not so.

[A]

Okay, so basically, I. I read the like. Or whatever, which I guess they don't know how to, like, justify text because it's like, just a really tiny margin.

[B]

But anyway, they got media.

[A]

I don't. I don't know. I really don't know. Like, I read this. I'm just like, I. So basically, what is the point of this? It just feels like a salty. Like, I don't know. I really don't know what the goal or point of this is.

[C]

Where's the pricing page? Find it.

[F]

Yeah, the merch. The merch is coming. All right.

[C]

Oh, yeah.

[A]

I really just like. They're like, we. We don't like things that no one else likes. Like. Like privacy. We care about privacy.

[C]

It's like, okay, well, that's a manifesto.

[A]

So we. It's a manifesto, but it doesn't have any manifesting in it. It's just like, I don't know.

[B]

Look good on them if you want to start up your own. If they call themselves anything else than restarting the Cult of the Dead Cow, then no one would give a shit. So good on you for pulling up a name for something that meant something back in 1999 and 2000. I guess.

[A]

I don't super understand this. It's basically just like, the Internet isn't free. Shucks. All right, see you later.

[E]

Why does everybody want everything to be free? Right? Like, I don't.

[C]

Kids these days, man. It's because of that phone.

[A]

When you go to the grocery store,

[E]

you're not like, why is this not free?

[A]

God damn. Like, okay, truthfully, if you're in the Cult of the Dead Cow, get in touch because what I want to know. I want to know one thing. What was the final straw? Was it AI generated Spider man videos? Like, what did it for you?

[B]

Was it surveillance capitalism? Is that what finally did it?

[A]

No, that's a minor thing, John. We've been doing that.

[B]

Yeah, that's it.

[C]

It was the spaghetti eating videos.

[A]

Yeah, maybe it was the spaghetti. I want to know, like, what was the final straw? Was it like having to tip 20% minimum at coffee shops? Is that what did. Like, I don't know what it was.

[C]

That would do it for me.

[B]

If a shit Gets back on a road to come back to Rage against the Machine. I'll be there. Then I'll be like, it's maybe that 99 again. Let's do this. But reviving cold to the dead cow, at this point, just start yourself. Start yourself something new. Don't build it on the ashes or something.

[A]

Also, yeah, a lot of those cults

[F]

of the living cast, just for the

[A]

record, a lot of those old school security dudes are some of the most toxic on the planet. But anyway.

[B]

Hey, hey.

[A]

He's in the.

[B]

John, you're fine, you're fine.

[A]

He's right, John. You've built an entire community centered around this fact. Anyway,

[B]

fine.

[A]

You know, so I think it's time to pivot and just talk about AI for the rest of the show.

[E]

Oh, God. Now the downhill.

[C]

We can start it.

[A]

Okay.

[C]

We can start it very lightly. Okay, there's one more.

[A]

Apparently someone at NASA got fished.

[C]

I don't know the one they got the ships, man.

[F]

The one I read. No, I read this one. There's no link to it. I didn't see it in the news.

[C]

Yeah,

[F]

okay, so pretty much this. The. This guy was pretty good hacker. No, it was a fishing organization out of. Based out of China who worked for Chinese aerospace. Right. Like they actually tracked him all the way back down. Launched a pretty good fishing campaign for the last, like, since I think it was 2017. In the article. See, zoom in on that. I'm in gallery mode, so I can't even read it.

[A]

2017, 2017. And targeting professors, researchers and engineers.

[F]

When was emailing people, asking them for, hey, can you send me the plans to that one machine? I left them on my other email and boom.

[A]

Oopsie.

[F]

There you go. Got some secret stuff right off the bat. And they've been doing it for a while. They tracked it all the way back to one particular dude, which once again, hackers can't get away. Hopefully. This guy probably can't leave China now,

[A]

but not that he came before because he had state secrets.

[F]

That's a good point. Yeah, but they popped him on a bunch of different charges, right? And waiting him. I think he's on the FBI's most wanted list now too.

[A]

Do you think the first fish was like, how did you go to the moon? Please explain in detail.

[C]

They wanted that rocket dude.

[B]

You know how that starts is like, I think the world is flat. And then you start arguing and you're like, well, I run satellites. What kind of satellites? Well, you wouldn't know them. Of course you would say that because it's part of the conspiracy.

[C]

Screw you.

[B]

Here's the top secret classified satel just

[A]

to prove that's for War Thunder.

[F]

I was about to say

[A]

this is. Okay, the truth. Put that on right now. It turns out Sung Woo is just really addicted to War Thunder. So anyway, yeah, supposedly he was really

[F]

good at pretending to be. Like, there was very, very good Oent done on each individual that they fished down to like, their friend group and him masquerading as people who they're like, friends with.

[B]

That's the way you do it, man.

[F]

Right?

[B]

You play the guitar on spear fishing. All right.

[D]

You had to go there.

[A]

All right, what else we got? We got, there's an unk. There's some new unks6692, which is a social, another social engineering campaign. Basically, this is a Google or Mandian or whatever post basically running through a bunch of same thing. It's more. We're really seeing heavy, heavy abuse of Microsoft Teams. And we are also, you know, abusing Microsoft Teams every day. It's because people are used to living in that. It's basically replaced the phone. From my perspective, it like teams is the new version of a phishing phone call and lots of threat groups are abusing it. There's specific things, you know, people are contacting people through Microsoft Teams and then they download some interesting stuff, Auto hotkey, scheduled tasks. Like, it's all pretty basic, but again, good social engineering goes a long way. There's some fishing pages and all that stuff. So.

[B]

And the big thing from this, I was interviewed on this one earlier today on another thing. And the thing about this is there's nothing really super, like cutting edge about this. It's just the speed of which they did it and all the different techniques that they chained together. And there was a lot of conversation of are they using AI? And I'm at the point now, or if anybody's asking if any threat actors are using AI, I'm like, yes, yes, they are. I, I, I don't I, it, yeah, it, it, it's like whenever we test things at bhis, they're like, whoa, would the Russians use this technique? Yes, yes, they would. I, it's just, of course they're using AI. Of course different groups are using whatever vulnerabilities and techniques that they can utilize. But it is interesting because we are actually seeing these threat actors change, chain these vulnerabilities together much, much, much faster than I think we have in the past. And Wade, I'd like to get your take on that too, because you have a lot of threat actors that kind of get stuck in one gear, use a handful of techniques. But this, there's a lot of techniques in this particular.

[A]

There's not any AI technology in here at all. This is old school.

[B]

My problem is we use AI too, for some of the stuff. It's not that there's an AI technique, it's the fact that they're chaining it together very, very, very quickly that leads me to believe that they might be using some AI with it. But that was my take.

[F]

Wait, I didn't read that one, but I believe you.

[B]

I'll go ahead and then.

[D]

Because.

[F]

Go ahead and go, Aiden.

[C]

Yeah, I mean, I think there still lends some credence to like what, what Corey is saying. I think it was Corey that said it, that there's a decent chance that they are still just doing like old school sort of stuff. But then to John's point, like you can script all of these things, but that becomes somewhat fragile. It would make a lot more sense to just have an AI orchestrate this sort of thing. Like it doesn't take a whole lot of effort to tell it. This is the, the chain of events that I want to have occur. And then I want you to ensure that, you know, if the output is this, go do that thing. If the output is that, go pivot this way. And I mean, that would be like an afternoon.

[F]

Yeah, yeah, I wouldn't even say that.

[A]

I don't know, I just feel like it's Google. If they could have pulled the AI lever, they would have because that's what it's hot to do right now.

[C]

But I mean, that's, you know, you

[F]

know what I'm kind of ready for is like, you know how, remember like, like living off the land bins and stuff, like when is it going to become lol AI living off different agents on their system.

[C]

And you say that we have actually an EDR agent as part of our soc that we can deploy to like a CLAUDE code session and so we can receive logs and then inject like EDR response into CLAUDE sessions. Like Ethan, one of the guys that works with us, we were testing it at one point at one of the Wild west hack infests where we were trying to block basically public pushes to like a gist or something like that. And so he goes to try to push something to a gist and basically the EDR agent just destroyed his CLAUDE session. Like he got a big warning banner and stuff. And we have more and more customers asking Us, like how do we monitor our AI? And I think that that's mostly just for like insider issues and people leaking things and stuff.

[E]

Stuff.

[C]

But I think you do have like a point is like that just becomes like a tool that's internal that this attacker can use. Right.

[F]

Well, think about a lot of the, A lot of the AI tools they're logging, one sucks. Right. And the other, they're just completely masquerading as a user. So you can't tell the difference between an AI doing something to the user, especially if they're using some type of OAuth or MCP on the host.

[B]

Right.

[F]

Don't even get me started. If they like bring in their own cloud code session and then they're on their computer and OAUTH to all your tools.

[A]

Right.

[F]

Like that you can't really see that.

[A]

And then topic is also everything because the server maintains the sessions, not the client.

[C]

Wait, we got to talk about that Claude code agent thing. I think you'd really be interested in that.

[F]

Yeah, there's, there's a bunch of fun stuff around. I'm playing around with some things, but we've been, we've been meaning to schedule a meeting for quite some time now.

[C]

Yeah, that's true. We got.

[A]

Yeah, just go on his pricing page. Just go on his pricing page. Purchase one, 30 minute, I'll pay. He's gonna vary. The pricing might vary depending on his mood at the time. So just keep that in mind. So continuing on AI, I think the bigger article that probably is on everyone's radar is basically a group is claiming to have access accessed Mythos. For those that have been living under a rock, Mythos is their new anthropic model. It's the new anthropic model that's so hot and spicy and dangerous that it's too powerful to be released. It'll never be released. But also apparently was accessible through some random API keys that some threat actors were able to compromise. So basically the like the summary here, which is generated by Apple and is completely invalid, we're gonna have to start over because it's Apple is basically that people claim to access it. They, the anthropic like group appears to have tracked it to some vendor API keys that were compromised. Kind of like a supply chain type API key compromised. So essentially at this point we can assume they didn't access them. They could access the model so they could make queries and they could do stuff. Anthropic knows exactly what they did with the model and when they did it. Like they already know this. It's more about, like, stunt, you know, stun hacking than it is actual impact. Because the truth is being able to run queries for a couple days through Mythos is not going to get you what you need. It's the actual model itself that you would need to really do any damage. And that was never at.

[C]

At stake.

[A]

But it is kind of just like a bad example for Anthropic.

[C]

Well, doesn't this say how is this.

[E]

Yeah.

[B]

Was this surprising at all? Right. Like, it was invite only. They gave. I don't know how many people they gave the invites to and I think they said initial cohort.

[A]

They didn't say. I mean, they're roughly 40 companies though. But I mean, again, it's like right

[B]

at that particular point, right? Like you have 40 companies. You can pretty much guess at least one of them is compromised or is going to do or at least to do something stupid.

[E]

They're compromised right now.

[A]

They should have done Mythos before they freaking published it.

[F]

The. I thought the article stated that they pretty much just guessed where it was. Almost like it was URL somewhere.

[A]

Yes, correct. That's right. They had to have the keys to access it, but then they also. Yes. Guessed where it would live and how it would work. Yes.

[C]

Well, part of this that almost makes it like way more boring is it says specifically in one of these sentences, one of the members of the group already had privileged access, access as a worker at a third party contractor for Anthropic, which to me means like, okay, like, there you go. That's how you got in because you worked with them like that.

[A]

You're not going to believe this bank robbery, dude. You go and you get a job at the bank every day.

[E]

They pay you from them every day.

[B]

As soon as this drop, they're like, oh, thank God we're still on the news. Like,

[A]

Yes, yes, John's right.

[F]

We don't need.

[E]

We don't need the new anthropic model. OpenAI just released 5.5. It's better. Yes, it's better. Just ask them. I'm calling them.

[A]

Right. So, okay, yeah, so that's another news article for sure. So GPT55, which by the way, after mythos dropped OpenAI released GPT54 Cyber, which is supposed to be their initial Mythos competitor, which they said is not going to be 40 companies. It's going to be thousands of companies. They're still gatekeeping it, but they're not. They're gatekeeping it less. Oh, then after 5, 4, now they've released 5. 5, which according to the numbers, if you look at. If you scroll down to basically where it shows the benchmarks, which is all anyone cares about these days. How does it do on Cyber Gym? It gets an 84 on cyber gym, which is similar to Mythos. And basically that's their.

[E]

Yeah. So they already released a Mythos caliber AI why is everyone still talking about Mythos like it's the second coming of Cyber?

[F]

Let me send you your. Your bank account real quick, Ralph.

[A]

I will tell you

[B]

the name. Like, you know, whenever we were getting these numbers, like five. Five.

[E]

Yeah. You're gonna change a whole new name because the whole new transition.

[B]

When you call something Mythos, I think that that is a much stronger pneumatic device.

[C]

Yeah.

[E]

Marketing right there.

[A]

Well, it's because they were first. They were the first also they, I will say, like, to hype it. They were the first to hype it. And to their credit, Firefox did confirm the bugs. OpenBSD did confirm the bugs. Like, when they reported the stuff, the companies that they reported on were like, yeah, this was big. This like Firefox got more bugs reported from Mythos than they had in the last entire year worth of Firefox article.

[B]

Because I. I got a bone to pick with that article. Like, finally the defenders are ahead of the curve and finally weird Defenders will be able.

[A]

They're about to announce their acquisition by Anthropic. Dude.

[C]

They're just looking for more money. They're like, please don't you want to bet?

[A]

Else you want to bet?

[B]

I can't remember. There was like 270 vulnerabilities that had discovered 180 some exploits, I think. And then Firefox is like, this proves as defenders, we're going to finally get it. It's like, no, we're effed. Like, as defenders in the industry, we are screwed.

[A]

Yeah. Okay, so dove. Dovetailing into us all being screwed. Okay, so there. This is the first. Like, I don't know how major this is, but. But essentially the Indian government published this kind of advisory that's basically just like, it's. I mean, it's kind of inflammatory, but it's also like, kind of true. I'll. I'll paste it here. Basically, they essentially said what John said on the news last week, which is every CVE is now plus one. Plus one.

[E]

Our podcast.

[A]

Yeah, basically the Indian government officially posted essentially the. The summary is like. Like anything you thought would get exploited in weeks will now get exploited in hours. Like, that's basically the gist of it. And so this is kind of interesting to see this on an official channel. Obviously, their website looks like it was vibe coded in 1996, but this.

[E]

This, you can't tell was not vibe coded because they know Vibe coded can do. At least that's true.

[C]

Actually, somebody drew that on a whiteboard before they coded that.

[E]

I think they coded it on a whiteboard and they had to transition it

[A]

to, I love it.

[B]

You guys quit on it. It's amazing.

[A]

Basically, it just says a lot of the things that John said last week. It pretty much just says, like, vulnerabilities can be changed by AI incredibly easily. Every vulnerability that would have taken weeks to develop an exploit now takes hours to develop an exploit. Like, you just have to kind of like it says right there, sharply reduce the time taken to apply patches within 24 hours is now the benchmark.

[C]

And you're talking about chaining a lot of these models, like, especially with 5.5. One of the biggest points that they tried to make about it is that it can do things autonomously for longer. And then Claude released their auto mode, which allows it to do things autonomously for longer. And so the benchmarks, like, whatever, they're going to be what they are, and they're always going to be better than your competitor in some way. Right. But the biggest thing that they, like, sort of like try to scoot in there under the radar in a lot of cases is our. Our model can go do what you've asked it longer before you actually have to do anything. And a lot of cases, it's true, like, 5.5 can do things longer and better than 5.4 can.

[F]

Can.

[C]

And then opus. Now, it'll usually work for a good long time before it actually needs something. Some of the times more recently, it will deliver these things to me broken. But it can do most of the work for a good extended period until 5.0 comes out.

[E]

It's going to be even better.

[A]

You got to go 4, 9 first.

[F]

Well, it's been taking longer for me.

[C]

Like, the OpenAI namings are like, chat GPT 5, 4, spark light, or like, whatever it is, if you get, like, under the hood, they're ridiculous.

[F]

You can't beat Mythos. It's just too cool. Like, we can do Odysseus.

[E]

They're never gonna release it, though.

[C]

They're gonna release it, but they're not

[E]

going to call it Mythos. They're just going to keep it as, like a.

[A]

They're going to call it Opus 48 Cyber.

[C]

Yeah, exactly.

[B]

I think the only thing that we can take from all of this is that Silicon Valley needs to come back. Like, that'll show. Like.

[A]

Yes, you've been saying this and I fully agree. We need to get like a reboot of like, it has to be a new show with new characters and new cast.

[B]

But no, we need to.

[E]

It's a whole new Silicon Valley right now.

[A]

So. Okay, stepping into some of the themes just quickly. There is this new trend of token maxing, basically, which is like it hit my radar diet technique. Yeah, exactly. So basically, essentially last week there was a startup with four employees that like bragged on Twitter about having an AI bill of $113,000 last month.

[D]

Month.

[A]

So like that's the economy that we're in right now. And by the way, the company that's on the receiving end is still losing money.

[B]

Drock. How much have we spent this month on AI.

[E]

Though? It's not, it's. If it's not big enough, that means that you're not using enough of it, which means that you're not going to succeed.

[A]

So anyone who would brag about how much they're using versus what they're actually making is not a good company. Yeah, tokens. Yes, exactly. I can have it analyze birds and burn a thousand dollars worth of usage for 10 months or whatever.

[C]

Speaking of burning a ton of usage, there is an article in here from like Anthropic's website where they did a marketplace for AI to basically negotiate things between their users. And so like it's, it's a silly project, but there's like two nuggets in there that are actually somewhat interesting if you like dig for them. But effectively these people in Anthropic like signed up up and they would like put up something for sale and the AI would negotiate with each other. So hey, I want to sell, you know, 20 ping pong balls or whatever. And then the AI would decide to price it and someone else's model would try to negotiate to buy that thing.

[A]

And so like, so it's Facebook Marketplace. I know what I got. Is Facebook available? Pretty much available.

[C]

The two like really interesting points. No, low balls were like number one, they said that agent quality does make a difference difference that the smarter agents got better deals. But then they also followed that up with people that were using or had like the dumber agents didn't realize it either. Which I keep thinking about because when, when I went to, I went to Japan a month or two ago and my wife was doing a lot of planning with, like, Gemini, and I realized she was on the, like, the free Gemini. And I was, like, wondering, like, how many times is it going to confidently say something incorrectly? As we get more and more complicated into transportation schedules just because it's, like, 31 light or whatever, like, you could

[A]

absolutely ride a bike from Tokyo to Osaka.

[E]

Exactly.

[C]

It's not even that far.

[E]

Most people do it.

[C]

There was once or twice that we, like, had this weird convoluted train thing where you have to buy, like, two different tickets and you have to, like, scan a bunch of stuff, and it got that wrong. And I, like. I think that goes to, like, sort of make the point that a lot of the folks that are, like, behind the trend on AI don't realize that they are. They just. They think they're. They're doing the stuff, but they're using.

[F]

I was about to say we got to reveal the real secret here. Like, our heaviest AI user probably on this. This podcast, does not give his wife access to his.

[A]

To his AI plan. Dude, I. I share my Claude Pro subscrib.

[E]

It's part of the wedding vows.

[A]

Let.

[C]

Let me defend myself.

[A]

She can now to have and to hold tokens until death.

[C]

She burns tokens to help order our groceries each week now. Okay, I'll just say that that's it.

[F]

That's all she said.

[A]

Did you put her on Quinn or Min? No, that's five five now. Okay. You got to get her on Mythos, dude. You got to get her on Mythos.

[C]

I'll discover a vulnerability in Kroger. We're about to cook Kroger's API.

[A]

It's like, you just start getting free groceries. You're like, I am concerned about that. I asked Mythos to.

[E]

To find better deals on groceries, and now I just keep getting free groceries.

[A]

Well, why are they delivered by a guy with a hoodie and they have

[E]

someone else on the lie? I don't ask questions anymore.

[C]

You just do what he says. O, like, the big personality difference, though, between these models is, I think the biggest difference. Like, if you go from ChatGPT, like, 54 to 5 5, like, I had an agent that was, like, almost annoyingly trying to make quips. Like, I left it because it was so annoying that it was almost funny. And then it changed to 5 5, and it became like, a normal AI agent that would actually, like, keep a consistent tone instead of ending everything with a wise crack.

[E]

You were using X as AI. That's what it's.

[B]

Yeah.

[A]

So. And for those that are saying John Looks it's because he did ask Derek how much spend more money. How much we're spending.

[C]

He looked at the socks. AI Spence out of it for a little bit.

[B]

I'm sorry. I'll be fine.

[A]

We had mythos get us free groceries. There's cops coming. Don't worry about it.

[F]

Kroger Hayden Hayden's trying to write a new agent asking it stuff and then it's like, you know what? You don't look good in that mascara. And he's like, wait, what? Like his wife

[C]

it just for me.

[D]

Me.

[A]

Yeah. I think the whole mythos g like it's basically an arms race. And I think the thing, you know, maybe if we're predicting models for or models we're predicting articles for next week, I think some of the other players are going to enter this arms race, right? We're going to see China, we're going to see deep sea start enter this arms race. That's when it's going to get really dangerous, is when you start because then it becomes who can announce the most shocking thing they hacked with AI that's the next step. This.

[B]

But here's where this ends, right? None of these, like all of these companies with some type of supremacy in this space, right? They're all thinking, well we'll be the one. We're going to be the Google of AI that's where they want to be.

[E]

None of them Google want to be that AI Whatever.

[B]

Sorry, Google's out of the. We don't need to talk about.

[C]

They just invested a ton of money in anthropic.

[B]

But if we're looking at, if we're looking at where this ends, like none of this is patentable, right? Like their secret sauce is all mathematics that was written in the 50s and you know, 60s and 70s and some in the 80s. And when you're looking at all of this, where it ends is anybody will be able to go and pull down a model from hugging face that's going to do like let's say 85% of what the really big ones do. So if you're looking at investing in this and I want to ask you guys this, like, who would you invest in? Like who's going to be the one that's going to win? What is their end state? How do they win? Because if they start raising their prices then already like me at bhis, I was talking to Hayden. I'm like, dude, do you want me to get you something could put on your desk for $4,700.

[F]

Look at him.

[B]

He's Kitty.

[A]

Look how happy.

[C]

Yeah, like I've done that pricing. Like if you are trapped in an ecosystem, like you're cooked.

[A]

Yeah, we should go one by one because I have very hot takes. I'll go last. I'll let you go.

[B]

I'll let you go last. But for me as a business owner, if I'm looking at this and all of a sudden let's say they like double or quadruple their pricing so they can get profitable, then BHIS and all the people that we have here, we have the skills and we like effort. I'm going to spend a quarter million dollars, I'm going to build my own infrastructure and we'll just build it out and then we'll get small little units for the people that need about that level of power for the really high end stuff that we're going to build our own stuff. I don't see how they win. And like one of these guys makes a tremendous amount of ungodly amount of money moving forward. Unless you're Amazon or Microsoft or maybe even Oracle, who's running the infrastructure for this shit.

[C]

That's exactly what I was going to say.

[A]

Say, well, you got to be Nvidia, you got to be Amazon, you're going to be Nvidia, the platform. It's just like Facebook. Why do they win? Because they were the platform. The biggest winner out of all of

[B]

this is asml, who very few people know about asml, right. But Nvidia, asml, all of these different chip manufacturers, they're going to win. Amazon's going to win. I don't know how anthropic and you know, OpenAI and all these guys are going to win any of this shit. Maybe Google just incorporates it inside of Google and everybody's using it as part of Google searches.

[C]

Yep.

[B]

So I would like to know where's the end state who wins, who makes lots of money outside of the hardware and platform vendors?

[A]

By the way, we're not an investment advice. No, you're not taking

[B]

advice from us. You absolutely should, but you'll get what you heard.

[E]

Oh, all right. So I think that the winner here is totally nobody. No, the. So essentially what's going to happen is, is that we're still in the arms race right now. So I think that the buying the local device, we're still going to be in time, like for certain tasks it's going to work, but the arms race is still hot and they're going to invest as much money as they can to, to stay in front of it. So you want to use their model right now, but at the end is where we're looking at, like, that's what John was bringing up. Like, well, where. Where does it end and how does it look?

[C]

Right?

[E]

And honestly, it. It's kind of an explosion, right? So at that final point where you can't really go any further, and then it. It does break down the price, they're gonna have to find other ways to make money than just raising the price, right? And so, you know, depending on the model's capabilities and all these other things. And I think we're just kind of still in the early days right now. I wish that we could just take a personal model and run it and have everything. But right now, they're gonna throw another billion at the next model to get you to keep staying in the system for right now.

[F]

So, Wade, one of them reaches AGI, right? And then that makes them all their money back. Like, they don't even sell it. They just use it for everything else to make money. Stocks, you name it.

[E]

All right?

[F]

And then it.

[A]

The company becomes the AI, basically. Like, instead of Anthropic being a company, Anthropic is a model that has AGI and it operates on its own.

[F]

It nukes all the other AGIs right off the bat.

[A]

Basically. Snow crash. This is.

[B]

There you go. The one that destroys humanity wins.

[F]

Well, it doesn't necessarily need to destroy humanity, right? It just has to knock out all

[E]

the other AI Become slaves to the AI.

[F]

We're already slaves to the government, man. It's all right. Mine's just another. Maybe he'll be nice.

[A]

Good. Good job.

[B]

So let's do it.

[A]

All right, Aisling, what's your take One?

[D]

I think Xai is going to lose first.

[A]

I don't know who's actually going to hit.

[D]

I don't know who's actually going to hit AGI, but I buy the weirdly dystopian Larry Niven take on what AI AGI does. Once it exists, it ramps itself up until it understands what the hell is going to happen with the history of the universe, the future of the universe, and then it shuts itself off forever.

[E]

Forever.

[D]

It doesn't do anything for us. It doesn't try to take us over it naval gazes until it shuts down.

[B]

I like that one better than. Than Wade's.

[A]

Yeah, I mean, it's pretty dystopian in

[D]

that every time we try to make the machines do the stuff for us, it checks the out.

[B]

It feels like panda. When. When Po gets the thing, it's like, this is it. And it's like, yeah. The secret to. The secret to the universe is you make your own path and the universe is what you make it. It's like, that fucking sucks. And it's like, sorry, man, that's what all the billions of dollars built up to. So I'm gonna go do nothing now.

[A]

All right.

[E]

They found a meaning of life.

[C]

I mean, I mean, I'll. I'll approach it from like, a different angle. Is like, how do you win as like, the common person? Because I'm not. If I'm on my own dime, I'm not gonna go be able to buy an AI rig that can do the sorts of things that I could get from these other AI providers. I can't. It would cost 20, 30 grand, and then it would make my power bill go so high that the city would probably wonder what I'm doing. So, like, how do you win?

[B]

I know.

[C]

Yeah. How do you win as like a common person, right? Is you have to be, like, decoupled from these ecosystems. Like, you cannot be changed to whatever. Whatever provider is in front right now because they probably won't be in front in a week. Is. You need to be able to have things and prompts and skills that are transferable between these different models to be able to hop to whoever is currently the best.

[A]

Right.

[C]

Because Anthropic was top of the world for a while and now there's rumors of quality issues with some of their models and chat. GPT says, don't worry, we give you more usage and we're not having these same issues. So, like, you have to be decoupled in a way and you can do that through just changing subscriptions, just bill monthly. Or you can use like a Raycast, like you can. I could press command K right now and switch between several different models or API keys or whatever I want want. So I'm using the same platform just decoupled from whoever that provider is. Right. But I don't think there's like a. I don't think there's a good outcome for the common folk in, in this sort of arms race here because we will all get bled dry at some point.

[E]

Yeah. Yeah, I was. Oh, before you go, Corey, I was going to say one thing. I. I totally agree with that. In the sense that, like, if the AI is the level, like the level playing for everybody, there's somebody's going to want to outplay the.

[C]

The little guy.

[E]

Right? Like, if. If we're all fair now because AI, some. Something's Going to switch where only the rich have the really good stuff. Does that make sense?

[C]

So, yeah, like Mythos.

[E]

Yeah.

[A]

So okay, I disagree with that.

[B]

You barely keep tabs on the. They have. It's whatever good stuff is, it's gonna leak.

[C]

Yeah.

[A]

So basically. Okay, so my take, first of all, if I have to back a horse, I'm backing Google because they're the ones who are the best at monetizing anything, right? That's basically their entire business model. They're also the only company that does AI things that is profitable or even reasonably sustainable.

[C]

They only have everybody's data and emails.

[A]

If you, if you're looking at OpenAI hemorrhaging money like you wouldn't believe Anthropic arguably hemorrhaging money like you wouldn't believe Google not hemorrhaging money, actually doing fine. Basically, I think we're at a point like DoorDash was in and like 2017 where the only person that was actually benefiting was the customer. Because DoorDash was spending a lot of money to deflate prices. Restaurants were spending a lot of money to try to get DoorDash customers. The only people who had a good experience were the users, obviously. Now why is DoorDash so expensive? It's because they actually priced everything in accurately. AI is going to trend towards that. So like as example, I use Anthropic. I don't get ads in my prompts. I have friends that use OpenAI or use free chat GPT and they do get AI ads in their prompts. So like it's going to be the same thing of like Google is going to monetize their prompts. So it's kind of going to become the same thing as every other ecosystem where it's like how many ads are you willing to stomach and how much are you willing to pay? The more ads you're willing to stomach, the less you're going to have to pay. And that's a personal decision.

[C]

But, but to, to bring some like ray of light array of hope back into this. Right. Like John, John mentioned something earlier in the show about like some hardware that you can buy about five grand that can run a bunch of models. It's not going to run a frontline model. It's not even going to run like a sonnet. Right. But if you compare like and like, we'll. We'll just use haiku, like Anthropic's lowest, worst model. Right. This is still an extremely powerful AI model. And if you can run something comparatively to that locally and Just effectively on your desk you can do things that you know, you couldn't imagine having done several years ago. So I think that that sort of like technology will also scale up to meet that sort of changing landscape where right now we're kind of cooked. But as you know, more open source models get better. And there was some, you probably talked about it on the news but there was like a paper that came out from. I can't remember who it was but they did some crazy math and like really shrunk down a very powerful model. It's blanking on me now but like yeah, yeah, yeah, yeah, yeah, yeah, yeah. It was that Turbo.

[A]

So Turboquant is, is a Google basically technology that will shrink the value of the KV store. So that will significantly improve or reduce VRAM usage for home or for everyone, especially benefiting home users. I mean I think it's like self hosting and, and that stuff has always been the same. Like Google Drive came out and you could just upload whatever files you wanted to Google. Then nextcloud came out or what like, you know what I mean? It's like you have self hosted alternatives that have always been an option. AI will be one of those options. It will be, you know, the experience with Google Drive and with nextcloud are different and there's pros and cons to each.

[C]

I am, I feel like that's a fine, I feel like that's a fine comparison. But like if you look at storage costs over time, it took a really long time for those to become more affordable. And, and I don't know like if compute in terms of like GPU or CPU or even like VRAM power is going to get to that point where that's somewhat affordable in the near future. Like RAM prices just keep going up.

[E]

Well that's because they keep building data centers to keep going in the arms race. So we're like in the middle of it, right? But I will say there are a couple, there are a couple local model stuff that is starting to move into that space, right. So one of them that I currently have is actually like transcribing. You know, if you want to transcribe your meetings or whatever, you have all these people that show up to the meeting, they all run AI. AI. Well you could do that locally on your computer now. Like the models and the computers are fast enough, you can have all of that local. So there is more that we're going to see where that AI essentially large language model gets moved onto the device to do those chores as opposed to needing to send it off to Anthropic to do that. Does that make sense? So we will see that more and more. But you know, also Anthropic is going to come up with a model that can do more and more. So.

[A]

Exactly. It's like, it's like running Crisis, right? It's like on some level did we get past the point where running Crisis is actually necessary? Like did anyone actually play through the whole campaign? Like, I don't know, they just benchmarked. Yeah. Like that's kind of where we're at with AI models. Like they're going to get so good that it's like who is using the full capacity of this model? Like how are you? Honestly, even now, I would argue it's to the point where the only way to use the full capacity of something like an Opus 5 or whatever of GPT 55 or is to basically give AI control over your entire life and all your projects. If you're willing to do that, go for it, man. But like that's I think a little bit of a jump and it's only going to get more capable and I'm going to use less and less of the capabilities, if that makes sense.

[C]

And that makes sense. But I think you're also like in the security minority of that kind of space where you have this security mindset and you've seen like all the bad things that can go wrong. Whereas even folks in the security world, see, see this can save me 40 hours a week on things that I don't have time to do. And the trade off to them is worth that potential risk. Right. Because they're assuming if we get hacked because of something that happened to Anthropic, that's an excuse that we could use.

[A]

True. I mean it's not even for me, it's not even the security implications, it's just the quality and runaway considerations for me. So like, like if you talk to AI enough, you're going to back it into a corner. Even to this day, day it's less and less happens. But you know, like it's going to get into a point where if it had autonomously made decisions and it would have designed something that I would not have been happy with or it would have given me code, like it would have written something in a way that I wouldn't be okay with. And so like if you take the human out of the loop, which is arguably what you need to maximize the capabilities of these models, I think the quality goes down and it just, then that becomes a loop where it just gets. It keeps reading its own shitty prompts and is like, well, I don't know what to do. I'm still stuck. Suck. But I'm gonna try again. I'm gonna get. I'm gonna refactor the code one more time and see if I can get that bug that I didn't get the first five times.

[C]

It's turning us all into pms. Like, the better project manager you are, the better your output will be because it's like a. It's like a good defined requirements. Yeah, exactly. You give it a good prd, like, and it'll actually go and do probably a pretty good job of what you're asking it for. It can go and execute on these things, and they.

[E]

They.

[C]

You know, they say thinking. Right, Right. But they can only think so much. Like they're not going to come up with novel solutions. So you, as the human being, need to drive the process to get an actual result.

[A]

Yeah. Yep. All right. John said he regrets starting this conversation about AI

[B]

God knows we wouldn't have gotten here without me, so.

[C]

Yeah, he just put in the chat. Corey, please stop.

[E]

Cory.

[A]

Cory. All right, exit. Thank you all for coming. May Mythos not crush your life.

[D]

Well, that was lovely.