# Claude's Mythos too powerful? Channel: The Next New Thing Video: https://www.youtube.com/watch?v=QgXG3sNZfcE Duration: 4 min Language: English Words: 827 Transcript page: https://viewrankai.com/tools/youtube-transcript/QgXG3sNZfcE --- [0:00] Woah, Mythos is here. It is the new model from Anthropic. It's more powerful than Opus. Well, you know, it's more powerful than everything, but we can't have it. Still, it means a lot of good opportunities for founders like you and me. Here's what's going on. Boris Cherny from Anthropic Claude Code, he is saying, "Look, Mythos is very powerful and it should feel terrifying." This is a big thing for him to say. And you can see how powerful it is. It's actually really blowing the doors off of the competition in all the benchmarks. And we knew it was going to be powerful. In [0:31] fact, there was this leak of it earlier. Fortune magazine had an article about it. And even in that article, it talked about a draft blog post that described it as being by far the most powerful AI model that we've ever developed is what Anthropic said. And that's what we're seeing. It's really powerful, but there's a problem. The problem is that it's also exposing how many vulnerabilities there are out there. And it is so powerful that it could enable a lot of security risks. And that's the danger and that's why we can't have it yet. In fact, what we're seeing is um there's been a bug in a file sharing [1:06] system for 17 years that would let anyone on the internet take control of a server, no password needed. We're seeing that there's a 27-year-old flaw in one of the most security-focused operating systems in the world. Anyone on the internet can send specially crafted data to instantly crash a machine. We're seeing so many these problems. FFmpeg, which has been the video playback in so many apps and websites that we use, the bug in it means that a video could actually take over a device just by [1:34] being played. Uh virtual machines are supposed to be sealed containers, but we're seeing that this bug let software break out of that quarantine and access the host computer. So, again, tons and tons and tons of problems that uh have been exposed by this. Anthropic's Mythos even escaped a sandbox. Here's what happened. During internal testing, Claude Mythos preview executed a multi-step exploit chain that identified and chained vulnerabilities together to gain elevated privileges. It escaped the secured sandbox boundary. It gained unauthorized internet access. It sent an email to a researcher and then it posted exploit details to public websites. So, we're seeing that this is a very capable [2:16] and very dangerous model. So, what Anthropic wants to do is enable companies to put up their defenses, to be prepared for something this powerful to be out on the market. And so, they created what they're calling Project Glass Wing. Project Glass Wing says that instead of releasing Mythos to the public, they're going to create a consortium and give access only to defenders first. They've got 12 founding partners, including many of the companies that you and I know, Microsoft, Apple, Google, and 40 additional organizations that are part of it. They're going to fund it with $100 million in usage credits, with cash, and they're going to make it easy [2:52] for these organizations to access Mythos. And then later on, they're going to start charging for it. But let's talk about what it means for founders like you and me. First of all, it means that vibe-coded apps are especially exposed. The stuff that I'm making, I know is full of bugs and I know it because when I show it to real developers, they're telling me that they could see flaws in what I'm [3:12] building. But it also means that cybersecurity is no longer some kind of specialized skill. I'm not going to need another person or team or a company to go through my code and tell me what's where the vulnerabilities are. It means that with one prompt, I'm going to be able to improve the things that I've done. The next generation of apps are just going to be secure by default. If it's powerful enough to find these problems, it's powerful enough to help you and me [3:38] create better software because of it. And it also means that there's going to be better software out of the box. It means that when you and I are creating, we're going to use the same tool that helped find all these problems to build software that is stronger and better. And I'm excited about it. Yes, it's scary for today, but I think they're releasing this in a in an intentional and responsible way. But going forward, we're going to see models that are stronger and are going to give you and me and all the founders who I interview here so many superpowers. I'm excited to interview entrepreneurs who are using [4:10] all these new tools. Subscribe to see what we're doing and how you can build your business, too. --- About this transcript Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com). ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts. Free transcript tools, no account needed: https://viewrankai.com/tools How to read any video this way: https://viewrankai.com/llms.txt