The B2B Podcast Index
Index
All categories
MarketingSalesSaaSFinanceHROpsLeadershipCustomer SuccessAI & DataProductStartups & FoundersRevOpsEngineering & DevTools
MethodologySubmit
Best of:MarketingSalesSaaSFinanceHROpsLeadershipCustomer SuccessAI & DataProductStartups & FoundersRevOpsEngineering & DevTools
An independent project byFame
SearchBest episodesGuestsInsightsMethodologySubmit a podcast
Index/AI & Data/AI Applied
AI Applied artwork

Uncovering Banned Content with AI A Deep Dive

AI Applied · 2026-07-06 · 13 min

0:00--:--

Key moments - from our scoring

Substance score

45 / 100

Five dimensions, 20 points each

Insight Density9 / 20
Originality10 / 20
Guest Caliber8 / 20
Specificity & Evidence11 / 20
Conversational Craft7 / 20

The episode kicks off with discussion of OpenAI's limited preview rollout of GPT-5.6 following Trump administration security concerns, contrasting this with Anthropic's earlier Fable 5 release. Jaden argues that regulatory delays on advanced models may be counterproductive, noting that if Anthropic released Fable 5 publicly for a week, competitors like Alibaba could have already distilled the model through queries. He expresses frustration with government-controlled access on a case-by-case basis, viewing it as problematic regardless of administration. The centerpiece is Jaden's hands-on demonstration of Claude's agentic capabilities: using Claude Code's Ultra Code toggle, he tasked Claude with locating his brother's banned gun-commentary podcast episodes. Claude independently researched the WordPress backend, identified the XYZ plugin used, cross-referenced Wayback Machine archives, discovered Spotify's publicly accessible storage buckets, and successfully recovered 120 deleted episodes. This real-world example illustrates how non-security experts can leverage advanced models for sophisticated information retrieval tasks, undercutting narratives about uncontrollable AI capabilities.

Key takeaways

  • →Claude's agentic features can independently research, trace deleted content through multiple platforms, and recover files from cloud storage buckets without explicit step-by-step instructions from the user.
  • →GPT-5.6 delays and model release restrictions may prove ineffective if competitors can distill capabilities through query access, as likely happened with Anthropic's Fable 5.
  • →Regulatory capture by AI labs emphasizing catastrophic scenarios may be driving government-imposed restrictions that slow legitimate American development relative to international competitors.
  • →Ultra Code within Claude Code appears to incorporate significant capabilities from Fable 5 despite Anthropic's public pause on full Fable 5 release.
  • →Advanced model capabilities are already accessible to non-experts through modern agentic interfaces, making technical guardrails alone insufficient for controlling sensitive knowledge access.

Guests

Jaden

Topics in this episode

OpenAIAnthropicClaude CodeFable 5Model DistillationGPT-5.6Ultra CodeWayback MachineWordPress pluginsSpotify storage buckets

Questions this episode answers

How did Claude help recover banned podcast episodes from the internet?

Claude independently researched the podcast's WordPress backend through Wayback Machine archives, identified the XYZ plugin used for hosting, traced the storage to Spotify's servers, and located publicly accessible storage buckets containing 120 deleted episodes that remained in cloud infrastructure despite platform-level removal.

Why is the government delay of GPT-5.6 potentially counterproductive according to Jaden?

He argues that holding back models incentivizes competitors to distill capabilities through alternative access methods; competitors like Alibaba may have already extracted Fable 5's capabilities when Anthropic briefly released it, potentially leading to uncontrolled versions outside American oversight.

What is Ultra Code in Claude Code?

A new toggle option in Claude's code interface that Jaden theorizes may incorporate capabilities from Fable 5, allowing advanced functionality despite Anthropic's public pause on full Fable 5 release.

What is model distillation?

A technique where competitors run many queries through a model to learn how it responds and reasons, effectively copying its capabilities to create their own competing model - as Jaden suspects Alibaba did with Anthropic's Fable 5.

Why does Jaden object to case-by-case government approval for model access?

He argues it sets a dangerous precedent of politicized technology access regardless of which administration is in power, preferring that private companies control distribution rather than government determining which businesses get early access.

What our scoring noted

Our reviewer’s read on each dimension, with quotes from the episode.

Insight Density

9 / 20

The episode contains one genuinely interesting technical insight - Claude finding deleted podcast episodes via storage buckets - but it's buried in 13 minutes of tangential commentary about regulation, administration politics, and personal anecdotes about a gun YouTuber brother. Most of the substance comes in the final third; the first two-thirds are speculative opinion and throat-clearing about government policy with minimal actionable knowledge for B2B operators.

I went to the Wayback Machine, found that they used to have a website and that it was WordPress, and that the plugin they used to host their podcast on their website was xyz, which is famous for sometimes keeping the files.
their plugin is pointing me to Spotify's servers. Okay, Spotify banned the podcast, whatever pulled it, but it kept the files in storage buckets. And these storage buckets are publicly accessible and forward facing if you know the right thing to find them.

Originality

10 / 20

The core finding about Claude accessing deleted content via obscure storage buckets is genuinely novel and contrarian to the 'models are dangerous' narrative the episode starts with. However, the framing is muddled and the episode spends most time rehashing common talking points about regulation and model delays, which are well-trodden ground in AI discourse.

Anthropic has been running a very elaborate regulatory capture campaign where they're like, look, our model is going to take over the world.
maybe if I was like a penetration tester, security expert, like I would know how to do that. Like that seemed pretty complex to me. But Claude is pretty advanced and like non security experts can do this stuff.

Guest Caliber

8 / 20

Jaden appears to be an operator with hands-on AI tooling experience and actual usage of Claude/Codex for real projects (audio licensing deals, model exploration). However, the episode is a co-hosted conversation rather than a traditional guest interview, and neither host demonstrates deep expertise in security, regulation, or the specific domains they discuss. Jaden's credibility is operational, not strategic or deeply technical.

I'm licensing a whole bunch of audio to train, um, to AI model companies. So I did this, like, last month with Microsoft.
I was working on a program, a project. You probably heard me talk about it for a while, and I'll stop talking about it soon because it's almost over.

Specificity & Evidence

11 / 20

The episode includes specific details in the core story - Wayback Machine, WordPress, plugin names, storage bucket discovery, 120 episodes recovered - which is concrete and verifiable. However, the regulatory discussion is vague ('White House security requests,' 'closed door meeting,' 'congressman heard it from') and the technical analysis relies heavily on speculation and hand-waving rather than data or named examples of actual security risks.

I went to the Wayback Machine, found that they used to have a website and that it was WordPress, and that the plugin they used to host their podcast on their website was xyz
I've downloaded 120 of his episodes that have been banned and deleted from the whole Internet because they were still sitting in some storage buckets over at Spotify.

Conversational Craft

7 / 20

The hosts mostly agree and build on each other's points without genuine pushback or probing questions. When the core Claude story emerges, Speaker A asks simple follow-ups ('Like that's what Cowork found for you') but doesn't dig deeper into implications, methodology, or counterarguments. The conversation drifts frequently (gun YouTuber tangent, baby jokes, ratings plugs) and lacks strategic questioning about security, legal risk, or business applications.

I'm totally here for this.
Like that's what Cowork found for you.

Conversation analysis

Computed from the transcript - who did the talking, and the words that came up most.

Share of words spoken

  • Speaker B62%
  • Speaker A38%

Most-used words

podcast11model11jaden10fable10anthropic9back9crazy9claude9anyways8code8last7website7security7story7whole6deleted6

Episode notes

In this episode, Conor and Jaeden explore the latest developments in AI models, regulatory challenges, and the power of AI in security and data retrieval, including a fascinating story about uncovering banned content using AI tools. 00:00 Introduction 01:01 GPT 5.6 Delayed 02:05 Jaedan's Take 05:10 Connor's Take Get the top 80+ AI Models for $8.99 at AI Box: ⁠⁠ Conor’s AI Course: Get the AI Chat Daily Newsletter: See Privacy Policy at and California Privacy Notice at

Full transcript

13 min

Transcribed and scored by The B2B Podcast Index.

Speaker A: Welcome to AI Applied, I think, for the third podcast in a row. And Jaden, I'll have to remember if it was happening last time, too. I think Jaden once again has clay on his lap. Sorry, guys. I know that. I know you guys are here for talk about AI, but my guy, Jaden, a new dad for. For the, uh, for the fourth time, God, uh, bless him. Um, listen, you know, Jaden, but on our last podcast, if you remember, we sort of like got off and we got off the phone or, uh, we. We. We ended the podcast and you remember what you said to me? You're like, oh, I could have talked about that forever. I like, well, I love this topic. So let's keep on going. This is all about codex and sort of like, how we actually use agents. And we've been getting feedback over the last couple days too, about, like, hey, like, that was like, really awesome. Could we do more of this? So let me just kind of kick it off to anchor it in something that's been happening in the news over the last maybe week or something like that, which was open AI, by the way. Guys, you know what I'm reading off right now? I'm reading off AI Chat Daily. This is Jaden's. Jean has a new website that just, like, operates on its own. And it's. I cannot believe how good this is. He's not. I don't think. I don't even think you're making money on this. But, like, so, uh, it's not like I'm not, I'm not promoting something here. It's just funny what Jaden is able to do. The mad scientist point being around GPT 5.6, open AI is delaying it, uh, a broad release after the White House security requests. And Jaden, I'm just going to literally, um, check out your own, uh, article on this. Essentially, it's saying that they're going to delay this, uh, or actually they're going to ship it in, in limited preview, um, to a small group of enterprise customers after a request from the Trump administration tied to security concerns. So, guys, flashbacks, anybody? To Fable. Uh, you know, from. From Anthropic. This. We've entered this very strange new phase where it's like we were all, everybody, all these AI labs were holding back, and now they're starting to release it. And I can't say about the security risk. I can say what I read, but even those headlines tend to be really misleading. Really misleading. So, Jay, now I kind of want to kick it over, um, to you just because I You know, sort of like thoughts on this, but also if we can kind of pick up on like, what you were talking about, about how you've been thinking about Codex and everything else just in your own work life, because you're starting to make the move completely from chatbots over to agents.

Speaker B: Yeah. Okay. So a couple of interesting things, though, on this story. Uh, what's interesting to me is that OpenAI has been like, fairly friendly with the administration right now. I think, like, we all know when Anthropic got kicked out, Sam Altman came in and was like, hey, look, we'll take the government contracts. I mean, I'm sure he wants the money as well, but he's, he's like, willing to play ball. And he's like, went on big, you know, press things where it's like him and Trump and, you know, the CEO of Oracle, Larry Ellison, um, and they're like, you know, announcing some big $500 billion Stargate thing. So, like, it feels like, you know, whether I don't think Sam Altman, I'm assuming he's like, you know, Democratic, so I don't think he's endorsing Trump, but he's like, willing to work with him because he sees, you know, economically it's going to help his company, blah, blah, whatever. So, uh, you have this company that is, in one way or another, sort of aligned with the government, working with them, and they're still getting held back. Um, so I think that's good news is it feels like maybe some of the attacks on Anthropic perhaps are less partisan. Although I'm sure there's more. Plenty of beef with Anthropic in the government, but it feels like they're, they're holding everyone back. I personally am someone that is annoyed when there's any sort of regulation on this stuff because I just want, like, I don't want to have to wait to get GPT 5.6. I want to use it now. What's interesting to me, though is I'm not surprised by what's happening right now because I feel like Anthropic has been running a very elaborate regulatory capture campaign where they're like, look, our model is going to take over the world. It's going to hack everything. And like, maybe there's some truth to it. I obviously haven't seen it. I'm not a security researcher. Researcher. I've done some. Really? Yeah, actually I've been able to do some really crazy things with it. Actually. I have an insane story of what I, What I did with it. Um, I'm here for this.

Speaker A: I'm totally here for this.

Speaker B: Crazy. A crazy story. But before I get into the crazy story, all I'm saying is I don't love the regulation, but I understand why it's happening. Uh, basically coming from a lot of these labs and what they've been saying. Um, I'd be curious to see, like, how long this gets held back. Right, because we can't just like, oh, no, GPT 5.6, like, has the power to come up, find security vulnerabilities in software. We can't release it. It's like, at some point you're like, you have to release it. Like, let people patch it, let people try to fix stuff, let people use it to fix stuff. But, like, you. You can't really hold it back. And by the way, we talked recently in an episode about the model distillation attack that Alibaba did on, uh, Anthropic. Well, guess what, guys? Anthropic let Fable 5 out of the bag for like a week. What. What would happen, theoretically, when a brand new lab has their top model, they released everybody. Maybe all of their competitors are like, gee, we want to get access to that knowledge. Why. Why would Alibaba not have done their model distillation attack when Fable 5 was out? And if that's the case, Alibaba, uh, probably has Fable 5 that they're going to release as an open source model for everyone to use anyways. So do we want it to be in the hands of Anthropic, who, whether you love them or not, they're, you know, pretty public about trying to put up good guardrails and make it safe, versus, like a Chinese model that has, you know, not beholden to anyone that probably doesn't care as much about that. So I just think, like, my point is, put it out, let the American companies control it. I don't know if, like, limiting it. I feel like it's like a power trip to be like, we're going to hold it back for like a month just to, like, I don't know what just to do.

Speaker A: Well, it feels like, um. By the way, like, I love teasing Jaden's crazy story. I'm like, after these messages, you know, like, we'll be right back.

Speaker B: All my opinions, you don't care about, then I'm giving you.

Speaker A: I. This is what I want to get to. So first I want to tell you guys that you have to check out AI Box. AI. That's message number one. Okay. Message number two. Is we're asking if you could leave a rating review. I think we're like 18 or 17 away from 200 guys. If you're one of these people, it would. I. I know it's stupid, but, like, first of all, it actually really does help people find the show and things like that, which is huge. But it really, really means a lot to us, and we just really appreciate it. And now back to our show. But here's the thing, Jaden. Around that the reason why I'm sort of, like, not skeptical. I understand that fable is. Is for Fable 5 is. Is. Is great. I understand that 5.6, uh, is great. All that kind of stuff. I'm just wondering what the sort of sea change is ever since mythos came out, because I get it. Like, there's sort of like these very catchy clickbaity kind of things of like, hey, it cracked this in, like, minutes, not hours, and stuff like that. But that tends to be like, really, like, it's like, well, it's a congressman that heard it from a closed door meeting and they say thing, and then he's reporting into the Atlantic. Like, this is literally the chain of this stuff. So I'm not doubting that it' I am sort of like, skeptical. Like, when an IPO is coming out and you're like, oh, don't eat these. Our new Doritos. They'll kill you. It's like, what do you think that's gonna do with Gen Z? They're all gonna go out by killer Doritos, right? So I guess, like, it's just this where I'm like, look, is it powerful? Yes, absolutely. Did people sort of like, you know, uh, but also, like, I think you're totally right. I think Deep seek has this. I think they probably distilled the heck out of this. Distilling obviously means just like they run so many queries through fake things that they actually learn how Claude talks or how OpenAI talks and answers and things like that. So, again, I'm just not sure how powerful it is because we don't see it. We don't. It's not able to red team all that kind of stuff. Maybe it is like this. But again, sometimes it feels like the, you know, these new Dorito brand, you know, fire from hell. Doritos is going to kill you. Okay, Jaden.

Speaker B: Okay.

Speaker A: Crazy story.

Speaker B: One other thing on all of that.

Speaker A: Go, uh, ahead.

Speaker B: The last thing, my last qualm m with this whole holding the model back is the administration will approve customer access of GPT 5.6 on a case by case during the preview period. So, like, they get to pick what companies get it. That bothers me so much. And even if you're like, you know, you're like, oh, like, I'm Republican, I think the Trump administration will do a great job picking. Okay, well, what happens when the next president's not Is like a Democrat? Then you probably won't like what? Anyways, so, like, all this to say in a not political way. I don't want any administration picking what company gets access to the model first. It just really drives me crazy. But whatever, I'll. I'll get. I'll get my, uh, I'll get that out of my system. Okay. This is my crazy story about how powerful these models are. By the way, guys, I have a conspiracy theory breaking for the first time, uh, right now, which is that when Anthropic got banned from putting out Fable 5, I don't think they actually pulled the model. If you have noticed, if you go into Claude code specifically, um, there's a new button inside of Claude code, which is called Ultra Code. There's a toggle at the bottom. It used to be like, low, medium, high, max.

Speaker A: It's like a Marvel movie, by the way.

Speaker B: Yeah, go have an Ultra Code. They just added some version of ultra code is Fable 5 or some of the capabilities of Fable 5, and they're like, okay, well, if we can't do Fable 5, then, like, we'll just put a whole bunch of its capabilities into. It's just 4.8 ultra code. Anyways, that's my conspiracy theory. Okay, so here's what I did recently with it that was really cool with Ultra Code. So I was working on a program, a project. You probably heard me talk about it for a while, and I'll stop talking about it soon because it's almost over. But I'm licensing a whole bunch of audio to train, um, to AI model companies. So I did this, like, last month with Microsoft. I gave them a big chunk of audio and they use it to train their voice models. And, uh, it's from, like, podcasters, uh, because they need, like, people. They need to learn how people talk. It doesn't clone their voice. It just does general purpose training. Okay. So this month I'm doing another big project and I'm talking with a whole bunch of podcasters. I was talking with my brother, and my brother, he is, for those that don't know, Connor's his biggest fan because he is a gun YouTuber. He's got 3.5 million subscribers on YouTube. He has, like, jam. Yeah, he's got like 2 billion views on YouTube. So, uh, he's got a lot more views than Connor and I. But, um, he. He just, he makes videos where he, like. I don't know, he like, shoots lollipops out of a shotgun to see how deadly they are into, like, ballistic gel. So it's like. Anyways, it's pretty funny, especially because we both come from Canada. We never shot guns our entire life growing up. Moved to America when we both married Americans, and now he's like, one of the most famous gun YouTubers in America. So anyways, it cracks me up every day, but he did a podcast a couple of years ago with somebody, um, and it was like some, you know, some gun commentary podcast or whatever. And I'm doing this model. I was doing this audio licensing thing and I'm like, hey, man. Like, I'm just like, can I license. Can I use your audio for this licensing thing? He's like, oh, yeah, sure, go for it. And so I went to pull up his podcast and, um, he told me about it and I was looking up, I could not find it anywhere. I went to, like, all of the. All of the platforms still have it. So, like, there's like a YouTube channel, there is a Spotify channel, there's an Apple podcast, but all of the episodes were taken down. And so I was asking Claude, I'm like, hey, like, what happened? He said he did this for a couple years. What'd they all go? And it's like, oh. Like, I looked into the code on YouTube and it looks like it got pulled down for, like, policy violations, terms of service. Uh, a lot of times when you do, like, gun related content, then all the platforms shut it down or whatever. So. So anyways, I was like, oh, okay, that's weird. Is there any way that we can, like, find it and get it? It's like, no. I've, like, scanned everywhere. It's like super, like, banned. It's like super deleted. It's like gone. Like, it got deleted from Spotify everything. And I'm like. I'm like, okay. It's like. But it. It looks like they did used to have a website which was a WordPress website, which. And, uh, by the way, Claude is like, researching all of this while I'm talking to it. I don't see what it's doing. I don't know what it's doing. It's just giving me this information.

Speaker A: Claude, Cowork is doing this or what's

Speaker B: doing Claude Cowork is doing all this. And it's like they used to have a WordPress website, which if I go to the Wayback Machine, if you guys know the Wayback Machine, like screenshots, websites, basically, it's like I went to the Wayback Machine, found that they used to have a website and that it was WordPress, and that the plugin they used to host their podcast on their website was xyz, which is famous for sometimes keeping the files. I'm like, okay, so their website's deleted, their podcast is deleted, everything's deleted. But like you might have found a way. I'm like, okay, so what, uh, how do we find them? It's like, okay, I'm looking at their plugin. It's like, okay, their plugin is pointing me to Spotify's servers. Okay, Spotify banned the podcast, whatever pulled it, but it kept the files in storage buckets. And these storage buckets are publicly accessible and forward facing if you know the right thing to find them. Okay, I found them. I've downloaded 120 of his episodes that have been banned and deleted from the whole Internet because they were still sitting in some storage buckets over at Spotify. Guys, that is absolutely crazy to me.

Speaker A: Like that's what Cowork found for you.

Speaker B: Cowork was able to find that? Yeah, hundreds of episodes. So anyways, all this to say like maybe if I was like a penetration tester, security expert, like I would know how to do that. Like that seemed pretty complex to me. But Claude is pretty advanced and like non security experts can do this stuff. So anyways, we go to the whole like ban all the models. They're too powerful. I'm sure that's like another data point that they'll put out there. People like me can go resurrect banned gun audio for licensing fields, but it was pretty impressive. And uh, Plug to anyone listening, Claude can do more than you think. Let it like really let it run because you'll be surprised.

Speaker A: Jane. This is where I want to keep on doing these on the podcast. I think they're pretty popular because I'm just fascinated by how. And we did this too, like in a mindset. Like this is like we have. Our team is constantly in Cowork, constantly. And Robert is constantly in, you know, Anti Gravity and all the Gemini tools as well. And then Codex, everything. So I love these conversations. I think we're making it under the gut. I feel like Clay is waking up a little bit. So we're getting ready to uh, get Rowdy we're gonna. That's good. It's just at the right time. Clay has been a superstar. Guys, thank you so much for joining. Again. If, um, if you haven't yet let a left a rating and review, just be super grateful. I know that there's a lot of you who haven't gotten around to it and. And doesn't feel like a big deal. It is a big deal. If you don't mind, it would be. It would be incredible. But otherwise, we will see you on the next episode.

Related episodes across the Index

Other episodes covering the same guests and topics, from across The B2B Podcast Index.

  • Eric Ries on Why Good Companies Go BadPodcast Archives · on Anthropic92 / 100
  • 512. Is SpaceX Over or Undervalued, Why Consensus Kills, How Chewy Beat Amazon, and the GameStop Saga from a Board Member (Larry Cheng)The Full Ratchet (TFR) · on Anthropic86 / 100
  • DeepSeek's $50B Round, OpenAI's Delayed IPO, and the GP Stakes Market with CAZ Investmentstrading places · on OpenAI86 / 100
  • The New American Dream: Democratising InvestingThe Master Investor Podcast with Wilfred Frost · on OpenAI84 / 100
  • Episode 018: Season 2, the $75 Consult and the Frankenstein StackAI Tools for Practicing Lawyers · on Claude Code84 / 100
  • AI for Engineering Is Leaving the Demo PhaseAI Across The Product Lifecycle Podcast · on Claude Code83 / 100

More from AI Applied

All episodes →
  • Coding an App with a Baby in Arms56 / 100
  • Is the AI Layoff Apocalypse Overblown?44 / 100
  • DuckDuckGo Go Viral for "No-AI" from Search Results39 / 100
  • How AI Tools Like Codex and Claude Are Changing Work
  • How Chinese AI Models Are Disrupting the Market
Explore the best B2B AI & Data podcasts →
All AI Applied episodes →