All of which matches what we said. The company produces the strongest model of artificial intelligence she's ever given to people. The model is coded, analyzes documents, helps science, can work directly with very serious security. It's three days, and the U.S. government actually says, stop! This model cannot be given to in-- strange citizens even if they are inside the United States, even if they work inside the company that this model has created. Anthropic had to turn off Fable-5 and Mythos-5 for all. And here comes the most important thing. What happened is not about Claude's broken down by Anthropic. It's a moment when artificial intelligence stops being an annex. It starts to be considered strategic technology. Not really like chips or missiles, but in reality, as a very serious weapon, yes, as military development. And the government says there's a risk of national security. Anthropic says we didn't show good evidence, we were given only an oral explanation about the narrow way the defence was circumvented. And moreover, other systems have the same opportunities. So the question is getting tough. If Fable-5 is to be blocked, why not block all the others? And beyond that, it is not just Anthropic, it is about any business that builds products on artificial intelligence. The main argument here is not one model. The main argument, who now controls the most powerful artificial intelligence in the world - the market, company or the state? Hello, everyone! You're on ToTheMoon, technological news, Silicon Valley sites. I decided to write a very quick special episode, considering that we only had, uh, a sub-cate on Sunday, and we were talking about what Fable-5 model came out. And Fable-5 is, of course, a very serious model from Anthropic, and it's supposed to be, uh, something totally unique, totally relative to the rest. Although it is, it's considered that, yes, it's what it counts. And there's no clear evidence. Illar says we're waiting for a back shot, so the blow from OpenAI. But, uh, this model, which is not working just now, there, with a context window in a million token, is a model that has, uh, a completely different regime for software development.
to ensure, develop different applications, analyse vulnerability, analyse, and, different external systems. And by the way, using Anthropic Mythos, aaah, it's a very interesting story to make it more difficult, but to make it easier for officials there, scared of chatbot now, and now I'm gonna tell you this story, and I'm gonna have to look at the context. Yeah, there was a Glasswing project, and this is a project Anthropic, where Mythos is their system of searching for different security vulnerabilities or creating security vulnerabilities. Ah, so this Mythos was used to find vulnerabilities to critical software under this Glasswing project. And Anthropic wrote himself that Mythos found, and his preview found a thousand high-priority vulnerabilities, including vulnerability in large operating systems and browsers, yes. And in updating this Glasswing project, the company reported that they were testing-- well, that work was done with about five dozen partners. I'll remind you that you couldn't, like a normal person, get involved and use this model. This model was used by only the companies that were identified, and there were more than 10,000 vulnerable people, that is, more than 10,000 high and critical vulnerabilities they found this-- in this project. There is also an interesting project that they have analysed the open source of projects and ninety, even there is a number, ninety points six per cent of the verified discoveries in open source projects have been real, confirmed vulnerabilities in the sample Anthropic described, yes. And by the way, finding vulnerability has taken more time than rectifying this vulnerability. And Anthropic reported that the early--medium rectification of high or critical vulnerabilities took about two weeks. Oh, and, uh, they had this conclusion that a major shift in the market that was, that it was easier to find vulnerability, and that they still got super, very slow. So, let's get back to what happened at all, and what happened now.
If briefly, in the case of a veil, the 9th June Anthropic produces Fable-5 and Mythos-5, and Fable-5 becomes a mass version. It's clear that there's a huge amount of protection inside it. And Mythos-5, this security system, it's only available to proven cybersecurity partners, there, biology, they sound like that. Oh, of course we were talking about Fable 5. I used Fable-5 in person from the ninth, ninth, tenth, eleventh, twelfth June. Not enough, Anthropic has a function. They used to have a level of work called Max, and they had a level of work called Ultracode. And, uh, that's a level of attention to what-- what the system does, that's a level, that's what it is, that's what it's called, you know, like, an old man, right? I personally felt that model was very interesting, even though it was too early to judge anything. I'm on this model, I'm a little bit of a technician, some kind of volume of start-up, but just start. Why? Because the 12th evening, it's last night, I'm taking off the, uh, Saturday morning, Friday night, five twenty-one night at our local time, Eastern time in California. Anthropic receives government orders. And there is a basic history that foreign citizens have access to the platform to be stopped. In fact, as there are no systems of normal analysis to date, the practical effect of the company is not only to foreigners, but to all, to avoid violating the directive, is it? So in the Anthropic statement, the key-- by the way, about not breaking me, my friends started writing, you're missing or not? And everyone was sure it was gone, disappeared for foreign citizens, and in fact they shut down everything. But because I, for example, use a lot of different systems, and there's no system in these systems, for example, that doesn't know my status. I mean, at least there was no further check that you should have loaded a passport and where the system could see who you are. Maybe they'll release this test the other day, and maybe it'll be something else. So in the Anthropic statement, Anthropic gave this statement to her, in the Anthropic statement, the key word that not a foreign user but a foreign national.
And we've been dealing with this for the first time. But you've heard the ToTheMoon Channel a lot about it. I've been telling you about this very much for the past two weeks. And all people everywhere, I say that it is a model of Anthropic and OpenAI that these companies are beginning to restrict access to their systems not for the United States, not for Americans. It feels really bad. I mean, not because there's any sanctions, not because there's any economic or other barriers, it's their internal condition, that's what they're like. Although Anthropic says otherwise, yet still. Which means it's a foreign national, it's been a lot wider. It means that a foreign national anywhere, there is a foreign citizen outside the United States, inside the United States. Once again, it was stated in the statement that a foreign national who works at the company Anthropic itself was also mentioned, which is of course very interesting. And that's why the company couldn't just take it away, that button. And we know perfectly well that she, she had to understand these legal statuses. And that story is stopping yesterday. So Anthropic writes that the government has invoked national security authority. And, but the letter did not specify what happened, who attacked, what harm was shown and why the State decided that it was the Fable and Mythos model that the two models should be disconnected. And why the same standard does not apply, for example, to others. In particular, Anthropic in its document directly compares and says that there are GPT 5.5 chat. You're referring to the problems that happened in other models, right? And it is clear that the subject of artificial intelligence is of great importance, that is the quality of evidence. And if the model is real, real, actually opening up access to cyberthreats, then, well, if it allows access to cyberattacks by some kind, the State understands that it has an obligation to respond. But if there's a narrow example of the reason, and yet the State says there's a narrow example of hacking, yes, Jellbrick, which can be repeated.
The one that can be repeated in other public models in this case, it doesn't look like just some kind of chaotic administrative decision, which looks, frankly, like, me, like the pressure, of course, on me.
Anthropic or a situation where, uh, the State doesn't understand what to do. Because, perhaps, in the security system, U.S. security agencies have been informed that, for example, the system has been used by hackers or some States that are, well, imputed enemies to the United States. But what, in particular, is obviously the model that these people wanted to use. And Anthropic is of the view that the Government has responded to the possible circumvention of protection of this model. I mean, once again, they released Fable-5, and she has a very big defense, yes. And that they were supposed to have been protected. And in the usual language, well, if we're just talking, it means that users make a request so that the model does what its protective rules normally do not permit. By the way, the company insists that this is impossible because we spent thousands of hours together with a huge number of engineers, including British people, to test this round just like this. Models, yes. And Anthropic is saying that if it is a possible narrow scenario where models ask to read there, I don't know, some kind of code base and find or cure vulnerability, it doesn't mean that-- that's not what it is. Doesn't mean there's a problem. That's very important, very important to understand and important to understand. The company says that it looked at the demonstration, by the way, where the method was used to find a few known or not very serious vulnerabilities. And most importantly, Anthropic states that other publicly available models can also find these things again. I mean, this is the foundation of the whole thing that's standing inside. Of course, by the way, well, Anthropic describes, again, Fable-2 was not naked, that it was launched with even more serious protection than, for example, the Claude Opus 4.8. And Fable-5 should not have revealed these possibilities that she had, very serious, than that Opus 4.8. And because one case there is a change in the cyphers, and the market didn't react for nothing.
So Fable-5 looked exactly different. I think even this ultra, ultra code that came into Claude Code, yes, well, this level-level ageing function, they have a low-, medium-high-old ageing, then extra, has max, yes, and there is. This ultra code that came in. Well, obviously, it's an ageing, and it's still for the model now 4.8. That Fable-5, well, is a really bigger model. They're hers, so they've been very serious about her and tested. And as I said, with-- and they had stress testing the so-called red team. So it's a straight hard, serious test to make sure there's no real problem. An important piece, very honest in the Anthropic statement, that Anthropic does not say that Fable-5 cannot be circumvented. She says that there is no single model supplier with perfect continuity to circumvent the rules. And you, we can't do any differently to test it and find it differently, yes. And in fact, you can't promise that the model is impossible to deceive. You can build a lot of protection, multi-level protection, make the rounds more narrow or very expensive. And that's what Anthropic calls it a lot-- she calls it a term of multi-level protection. That's what's important to understand. I'll remind you that the Fable model was produced as the Mythos model class. This is the 30-day storage of requests and answers for trust and security purposes. Elnar and I were discussing it. I'll remind you it's thirty days of storage. They, they separately examine the data, have the possibility to analyse these data, have the opportunity to make further claims to other companies that used it. And Microsoft, for example, banned its employees in the early days of the Fable model from using this model because of the fact that 30 days of data are stored and so it can happen very well. Serious leaks, yes, because they are not models that were closed in the case at hand in the Enterprise, in a corporate segment from some external audits. Although logic is, all the external analysis is that the external auditors were in any way, yes. I don't quite understand this uniqueness because there's a code in OpenAI for thirty days, and in fact, he can still go to the lab if the system so requires.
I remember Elnar telling you in the last edition on Sunday that the model, and you can see it, you know, we've been discussing a lot of things that's about the whole use of this, that the Fable model, and Anthropic said that if someone started to abuse this model, that she was supposed to even start fooling that user. Actually, what happened is a very serious challenge, and Anthropic says that they are implementing the legitimate directive now that they, well, block all access, but they don't agree. And they're categorically disagreeing, and they'll be totally clear about it. And this, that a narrow potential over-protect should be the basis for revoking the commercial model that has been developed for a huge organization. So they're not really in agreement with that. There's no need for that. And they say that if this standard is applied to the entire industry, it will actually stop new best-page releases. I will recall once again that the U.S. government has recently promoted the law that all new models should be negotiated with the State. After that, it stopped. Sam Althman also spoke about this. He said he disagreed. The Government cancelled this, and the Government then issued a new comment or directive stating that, generally, they can send these models at the company ' s discretion. It's a little time, a few days, a new cool model of Anthropic. It's clear that Anthropic is trying to overthrow the OpenAI, and I think they've become very serious. They've gained an incredible pace in the run-up to especially IPO. And we know what happened to IPO SpaceX now over two trillions of dollars. Elon Musk was the first trillioner in the world. SpaceX is inside XAI. It's very important to understand that it's not about space. If anyone suddenly turns on a skate and not within the framework, without understanding the end of the artificial intelligence, that is, there's XAI inside SpaceX. Anthropic, OpenAI also prepares IPO. Of course they'll be on the IPO, it's not gonna be a trillion-dollar, yeah. And there's a very serious struggle. By the way, Anthropic is not against State control because Anthropic has spoken long ago, and recently they have issued a large whole document, we'll have a separate special probability that the State should have a special state.
The possibility of blocking the dangerous deployment of different models. And in the June package of such political proposals that the U.S. government made, well, when the campaign was going on in yu-- well, they wrote in their proposals that the most powerful models needed tests, Some independent assessment and such a legal opportunity to stop a dangerous release. Well, actually, what they're dealing with, right? I think they've offered the State to block the models. The state came and said, "We have the right to block a dangerous model, and we don't understand how to work with it." There must be some logic in that, yes. To understand what I'm telling you today about this untechnical architecture of this story, it is important to understand that Mythos Five and this class of these models is the most powerful versions for sensitive models. The directions that exist on planet Earth, such as cybersecurity, biology, medicine, scientific research. There is even, there is evidence that these models are accelerating, and incredibly accelerating the development of medical research. It's ten times. I've seen this number dozens of times. And Anthropic was generally called Fable Five. Well, they said it was Mythos class, and that this model was supposed to be dealing with long, complicated tasks. So, these are huge, big code projects, yes, a huge analyst of documents, a reversal of this huge analyst of documents. It's not a question of the context window, believe me, there's a million tokens, yes. This scientific task was to be addressed in a very large, serious way. So this model is made to have a very serious visual understanding, yes, inside. Well, it's clear that long scenarios of parallel work of different programs, agents, code, some triggers and everything. I mean, it's not gonna be a question, it's gonna be, you know, something super serious, yeah, it's incredibly serious. And the price, by the way, is, of course, the cost of this model and the cost of the study, it's incredibly serious. I've got one project on my last night, and I want to put in a special analysis of my chat in the GPT chat room.
And I've been ana-- analyzed a very large piece of Fable model. While I'm, uh, there's already, uh, I'm just a lot of limits on it and a lot of tokens. And the Fable model spent more than 12 or 13 million, I think I've got tokens, uh, I've had it for my ChatGPT. And she's been looking at the 30 percent of everything. So, at least the piece of analysis was made Fable, and the architecture of this decision was made, by the way, Fable. It's very interesting because I want this architecture, and, uh, to scramble, yeah, to scream and make it accessible, um, to make it accessible to everyone. And again, uh, I want, uh, the main point of the whole thing to tell you, that we're going to move in a new time, of course. And last night, I even more, I realized more seriously these restrictions that we'd have, especially when you had access to something, and then you took it away from you. Fable 5, she was released temporarily and said that she would be available to everyone before 22 June, and then she would probably be taken. Well, what's the plan? She'll be, uh, for the tokens. And the projects I'm dealing with, like, 12 million tokens, 30 percent of my system's analysis would cost about $300 to $1,000. It was a project evaluation of this type. It's just a token, the entrance, the weekend. Is there a thousand to three thousand dollars to analyze this infrastructure of the system? I'm ready. But it's clear that most people aren't willing to spend so much money on their ChatGPT analysis. What do you think about this? How do you see that? Where do you see? Oh, give us comments, comments, liki. Don't forget to sign on our channel, if it's the first time. Look at our graduations. We're going out three and more times a week. I'll see you in the next edition.