Jun 15, 2026 · 57 min · 12 segments
The US government just laid down the hammer, officially recalling frontier AI models Fable 5 and Mythos 5. In Episode 46 of the DTF Cyber Podcast, Damian, Troy, and Fern dive into the chaos. Did a…
Yeah, Fern, so I just wanna, you know, get rid of some miscommunication or misunderstandings.
A lot of people have had a lot of calls this weekend about this topic and, you know, they didn't shut down Anthropic.
You know, they just said that they couldn't allow foreign nationals to use these powerful new algorithms or these new AI models, sorry, the frontier models of Fable 5 and Mythos-5.
But Fable 5, which came out Tuesday of last week, is now offline, and we don't have any clue when that functionality is coming back.
That's within a week.
That's pretty quick.
So Troy, are you saying somebody actually jailbroke the guardrails that Anthropic had put in place?
They absolutely did.
They have a ton of, you know, guardrails in place and, uh, mechanisms to detect when this happens.
But there's been at least two different documented cases where somebody used a very sophisticated prompt to get around the guardrails and safety controls of the Fable 5 and turn it more into the Mythos that the Project Glasswing folks have access to, meaning that it can find zero days faster, and the key thing about Mythos that's really interesting is the chaining together of vulnerabilities, being able to find a low, medium, or high using other mechanisms of low, mediums, and highs, chain those together to find their way into your environment and to your crown jewels.
That's the beauty of the Mythos model, which is why it was so restricted to that Project Glasswing consortium.
So what we're saying here is any average citizen could use a sophisticated prompt to turn the Fable 5 consumer-facing model into this weaponized Mythos-level zero-day-finding machine.
That's interesting.
So okay, so let me ask you this because, I mean, what you say kinda clarifies a couple things for me.
But let's just assume that everyone's just AI, AI, AI, right? And everyone wants to use the latest and greatest tool.
Now they don't have access to it, but in a situation where people start, um, we'll say leaning into the latest models internally for corporate use, they're doing it as part of their SDLC, right? It's doing a lot of their code.
It's helping them with a lot of their, uh, work and workflows.
Now this seems like this is just a third-party risk issue in the sense that...
Or maybe would it be third-party risk, or would it be more so, um-
It'd be-
Yeah, Fern, so I just wanna, you know, get rid of some miscommunication or misunderstandings.
A lot of people have had a lot of calls this weekend about this topic and, you know, they didn't shut down Anthropic.
You know, they just said that they couldn't allow foreign nationals to use these powerful new algorithms or these new AI models, sorry, the frontier models of Fable 5 and Mythos-5.
But Fable 5, which came out Tuesday of last week, is now offline, and we don't have any clue when that functionality is coming back.
That's within a week.
That's pretty quick.
So Troy, are you saying somebody actually jailbroke the guardrails that Anthropic had put in place?
They absolutely did.
They have a ton of, you know, guardrails in place and, uh, mechanisms to detect when this happens.
But there's been at least two different documented cases where somebody used a very sophisticated prompt to get around the guardrails and safety controls of the Fable 5 and turn it more into the Mythos that the Project Glasswing folks have access to, meaning that it can find zero days faster, and the key thing about Mythos that's really interesting is the chaining together of vulnerabilities, being able to find a low, medium, or high using other mechanisms of low, mediums, and highs, chain those together to find their way into your environment and to your crown jewels.
That's the beauty of the Mythos model, which is why it was so restricted to that Project Glasswing consortium.
So what we're saying here is any average citizen could use a sophisticated prompt to turn the Fable 5 consumer-facing model into this weaponized Mythos-level zero-day-finding machine.
That's interesting.
So okay, so let me ask you this because, I mean, what you say kinda clarifies a couple things for me.
But let's just assume that everyone's just AI, AI, AI, right? And everyone wants to use the latest and greatest tool.
Now they don't have access to it, but in a situation where people start, um, we'll say leaning into the latest models internally for corporate use, they're doing it as part of their SDLC, right? It's doing a lot of their code.
It's helping them with a lot of their, uh, work and workflows.
Now this seems like this is just a third-party risk issue in the sense that...
Or maybe would it be third-party risk, or would it be more so, um-
It'd be-
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.