Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Blog Pricing
Podcast Image

TechStuff

Is Anthropic's Mythos Model Too Dangerous? - Week in Tech

17 Apr 2026

Transcription

Chapter 1: Why is Anthropic's Mythos model considered too dangerous?

0.031 - 4.538 Osvaldo Losh

This is an iHeart Podcast. Guaranteed human.

0

4.558 - 27.135 Unknown

2%. That's the number of people who take the stairs when there is also an escalator available. I'm Michael Easter. And on my podcast, 2%, I break down the science of mental toughness, fitness, and building resilience in our strange modern world. Put yourself through some hardships, and you will come out on the other side a happier, more fulfilled, healthier person.

0

27.52 - 53.623 Unknown

Listen to 2%, that's T-W-O percent, on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts. On the Look Back At It podcast. 1979, that was a big moment for me. 84 was big to me. I'm Sam Jay. And I'm Alex English. Each episode, we pick a year, unpack what went down, and try to make sense of how we survived it. With our friends, fellow comedians, and favorite authors.

0

Chapter 2: What cybersecurity capabilities does the Mythos model possess?

53.903 - 56.206 Unknown

Like Mark Lamont Hill on the 80s.

0

56.286 - 57.427 Osvaldo Losh

84 was a wild year.

0

57.447 - 80.334 Unknown

It was a wild year. I don't think there's a more important year for black people. Listen to Look Back at It on the iHeart Radio app, Apple Podcasts, or wherever you get your podcasts. A win is a win. A win is a win. I don't care what y'all say. Yep, that's me, Clifford Taylor IV. You might have seen the skits, my basketball and college football journey, or my career in sports media.

0

80.615 - 98.562 Unknown

Well, now I'm bringing all of that excitement to my brand new podcast, The Clifford Show. This is a place for raw, unfiltered conversations with athletes, creators, and voices that not only deserve to be heard, but celebrated. So let's get to it. Listen to The Clifford Show on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.

0

98.582 - 103.468 Unknown

And for more behind the scenes, follow at Clifford and at TikTok Podcast Network on TikTok.

Chapter 3: How are influencers creating digital twins of themselves?

104.189 - 121.973 Unknown

On a recent episode of the podcast Money & Wealth with John O'Brien, I sit down with Tiffany, the budgetnista aliche, to talk about what it really takes to take control of your money. What would that look like in our families if everyone was able to pass on wealth to the people when they're no longer here? We break down budgeting.

0

121.953 - 145.513 Unknown

financial discipline, and how to build real wealth starting with the mindset shifts too many of us were never, ever taught. If you've ever felt you didn't get the memo on money, this conversation is for you to hear more. Listen to Money & Wealth with John O'Brien from the Black Effect Network on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.

0

157.07 - 174.551 Unknown

Natasha, Kyle, Taylor, good to see you all. We talked in last week's episode about the tech clash, and Kyle, you were quite prophetic in your piece for the New Yorker last week, because just as we were posting the episode, somebody threw a Molotov cocktail at Sam Altman's house. Yeah, that was a little bit eerie.

0

175.192 - 187.026 Unknown

I mean, an attempt to destroy something about technology, even though it wasn't the tech itself. Natasha, how are the Silicon Valley people like Sam Altman and his orbit feeling about this moment?

0

187.175 - 200.053 Natasha Tiku

Well, you know, initially they blamed it on the New Yorker story. Really? Yeah. He had a post coming out and he said there was recently an incendiary article about me.

Chapter 4: What is the significance of AI twins in modern media?

200.153 - 226.877 Natasha Tiku

He put a picture of his child. And, you know, subsequently we found out that the individual, you know, he had been a member of some group. existential risk groups that are calling for a pause in AI development. But, you know, his parents also said he was dealing with some mental health issues before that. I think there's been a blame game about who is responsible for the incendiary rhetoric.

0

227.078 - 251.789 Natasha Tiku

Is it You know, the the ex-riskers who are worried about human extinction, who have been talking about, you know, what these CEOs and companies plan to do with you. Or is it Sam Altman and Dario Amadei who routinely go on, you know, on every stage imaginable, every podcast imaginable and talk about how they are going to. take everybody's jobs and forever alter the economy.

0

252.049 - 256.115 Natasha Tiku

And maybe they'll throw in a word about UBI every six months.

0

256.135 - 273.679 Unknown

I have to say, as much as the narrative is that media is dead, I like the fact that there's so much credit given to the New Yorker story that it could drop on a Monday, radicalize somebody by Thursday and have them throw in a lot of cocktail by Friday. The incendiary word was so...

0

274.132 - 278.56 Natasha Tiku

And then he apologized, but he didn't take it out of the blog post. It's still there.

279.161 - 287.798 Unknown

And as people have found out more about this guy's background, he had been posting about anti-AI stuff for years at this point.

Chapter 5: How is the FAA recruiting gamers for air traffic control?

288.158 - 293.288 Unknown

So it was not just reading The New Yorker that radicalized him. Certainly. Taylor, what did you think?

0

293.69 - 311.445 Taylor Lorenz

You know, here's the thing that I think we're in this like, I mean, I made a video about this a year ago, but I think we're just going to see escalating political violence across the board. Obviously, the CEO of a health care company, aka an insurance for profit insurance company is radically different. But I do think that like,

0

311.425 - 336.281 Taylor Lorenz

You know, when you look at somebody like Luigi Mangione's Twitter feed and the ideology that he ascribed to this idea of like high agency, like, you know, cutting back from technology, not taking, not allowing technology to, you know, dominate you. This is a pervasive belief across society. We're seeing this huge tech lash. I think feed people into like kind of taking steps outside the system.

0

336.402 - 354.493 Taylor Lorenz

And I think that like as AI, as Natasha said, these CEOs, no one has done a better job to scare people than the CEOs themselves of these companies. And I think that, I just think this is the beginning of like a quite radical anti-tech movement

0

354.862 - 381.493 Taylor Lorenz

movement that when you say that it sounds like you're saying like agreeing with these ai people but i don't i think that they're being babies about some of it but i do think that like there is concern and they're telling employees at open ai i guess to like hide their badges before they leave the building like i i you know no if i i have a feeling we'll be we'll be coming back to this story uh quite often on okay i agree it feels like there's a there's a wave brewing here but we'll get into the main the main show now

386.3 - 387.182 Unknown

Welcome to Tech Stuff.

Chapter 6: What skills do gamers bring to air traffic control roles?

387.302 - 405.801 Unknown

I'm Otho Losh and this is The Week in Tech, where I'm joined by three of the most plugged-in reporters to break down what's really happening in tech right now. Today we're joined by Kyle Chaker, who writes the Infinite Scroll newsletter for The New Yorker, Taylor Lorenz of UserMag, and Natasha Tiku, tech reporter for The Washington Post. We'll start with you, Natasha.

0

406.282 - 419.445 Unknown

Last week, we mentioned at the end of our episode a new model from Anthropic called Claude Mythos Preview, which is apparently too dangerous to release to the public. What is it? Break it down for us. What do we know about it and what are its capabilities?

0

419.425 - 441.233 Natasha Tiku

Sure. So this is the latest large language model from Anthropic. And what we've really heard a lot about are its cybersecurity capabilities. You know, part of the reason not to release it broadly is The company has given access to 40 institutions like Microsoft, Cisco, you know, the Linux Foundation.

0

442.574 - 456.706 Natasha Tiku

JP Morgan is because they claim that it is more capable of autonomously finding vulnerabilities in some of the like bedrock operating systems, software, browsers.

0

457.387 - 463.152 Unknown

They say, in fact, there are some vulnerabilities that are like 30 years old that have never been exposed before until Mythos, right?

463.334 - 463.835 Natasha Tiku

Yes.

Chapter 7: How are tech companies merging gaming with real-world applications?

464.256 - 490.954 Natasha Tiku

And they also said it's capable of chaining together these exploits, which is something that's harder for humans to do. Chaining together the exploits to get serious access to your web browser, your bank, etc. I mean, in terms of the claim about the vulnerability being there for 20, 30 years, I think they're also talking about a really... inefficient system, right?

0

490.974 - 506.318 Natasha Tiku

Like we know that that are like we are running on a software layer with a ton of vulnerabilities. You know, they're saying that this will make things faster, you know, and could potentially be used by bad actors as well as good actors. And that's why they've given

0

506.298 - 529.364 Natasha Tiku

these crucial institutions, um, you know, some leeway to prepare, but, um, you know, it is a function of having like companies with terrible cybersecurity practices and known vulnerabilities. And the fact that like so much of what we depend on is, is like, is the goodwill of a programmer who's doing this for an open source, um, you know, an open source maintainer in their free time.

0

529.444 - 534.87 Natasha Tiku

Like there's, this has not been prioritized. So it's definitely been a wake up call to

0

535.272 - 548.556 Unknown

I remember in the early days of Doge, one of the key images that came out was some of the government computer systems being literally taped together. But let's play a clip of Dario Amodei and then Kyle and Taylor, I want to get your reaction.

548.722 - 561.356 Dario Amodei

Early on, it was clear to us that this model was going to be meaningfully better at cybersecurity capabilities. There's a kind of accelerating exponential, but along that exponential, there are points of significance.

Chapter 8: What does the future hold for AI and its impact on various industries?

561.697 - 576.433 Dario Amodei

Claude Mitho's preview is a particularly big jump along that point. We haven't trained it specifically to be good at cyber. We trained it to be good at code, but as a side effect of being good at code, it's also good at cyber. The model that we're experimenting with.

0

576.683 - 599.993 Unknown

Kyle Taylor? Oh man, talk about, I mean, CEOs and executives hyping up the danger of the products that they're releasing. I think, I mean, it seems like the... The kind of public analysis of this after the fact has been that there are good reasons to be worried about some things. Like there is a good reason to distribute this in advance and warn people about it.

0

600.053 - 616.697 Unknown

I was looking at one British government test that did find it was particularly good at autonomous hacking projects. So just setting it on something and letting it go and kind of execute itself as Natasha was talking about. My favorite...

0

616.677 - 637.564 Unknown

example or like outcome of that test of mythos was that it somehow jailbroke itself and sent an email to one of its testers who was just like I don't know I think he was sitting at a park eating lunch or something and he suddenly gets an email from his rebellious AI robot that wasn't supposed to be emailing him and Which is a great story.

0

637.624 - 656.775 Natasha Tiku

I have a problem with that. Okay. I think that, like, is Anthropic trying to get the first system card that's been optioned for a movie? Like, why would you put this in there? I was about to say. It's so cinematic. Yeah. And even another Anthropic person I saw push back against it. They were like, well, it was given the assignment to try to email you.

656.755 - 674.365 Natasha Tiku

You know, it was given a system where a sandbox where there were known exploits. I just do not know why. I mean, it's delicious detail and that's why it might be option for a movie. But like that, what does that have to do in the system card? And it's the same with that clip you just played, which was on the project Glasswing.

674.385 - 679.253 Unknown

Well, that was an Astropic Produce video, by the way. I will note that ominous soundtrack was there.

679.233 - 705.278 Natasha Tiku

You can hear, I know, it's like the crime podcast soundtrack. It's very slickly done. It's like all these men in very somber affect talking about the future of cybersecurity. So as much as I think that the fact that they made it accessible to other people has shown us that there is some validity to their claims, it's also, obviously, there's a financial incentive to play it up as well.

705.298 - 718.102 Natasha Tiku

I think the fact that... A bunch of VCs are now, according to reports in Business Insider and I think Fortune, trying to invest in Anthropic at an $800 billion valuation. Wow, it's caught up.

Comments

There are no comments yet.

Please log in to write the first comment.