Anthropic has put Claude Mythos 5, the most capable model it has for this kind of work, behind its Claude Security scanner. The scanner connects to a code repository, hunts for vulnerabilities, rates how serious they are, and suggests a fix.
Here is the plain version for anyone running a small business site: this is not for you, and that is fine. The Mythos-powered scan is in public beta for Claude Enterprise customers only. Teams on the Pro, Max, or Team plans do not get it. It also works by connecting to a GitHub repository, which means it is aimed at organisations that write and maintain their own software. If your site was built for you, or built on off-the-shelf software with a theme and some plugins, there is no repository of yours for anything to scan.
So why does it matter enough to mention? Two reasons, both indirect.
First, Anthropic says it is working with cybersecurity vendors to build Mythos 5 into their products. The security tools that ordinary businesses actually use tend to inherit improvements like this eventually, without anyone announcing it to you. You would just find that your host or your security plugin got slightly better at catching things.
Second, Anthropic is putting $35 million in credits into a fund for finding and fixing vulnerabilities in open source software. Open source means software that anyone can use and inspect for free, and a great deal of the internet runs on it, including large parts of what sits underneath a typical small business website. If that money finds real bugs in widely used code, the fixes eventually arrive in your updates. Which only helps if you install them.
That is the honest takeaway. Better scanning at the top of the industry is good news, but it reaches you slowly and second-hand.
What to do: nothing here needs your attention today. The thing that does is boring and unchanged: make sure someone is applying updates to your site's software on a regular schedule, and that you know who that someone is. If the answer is "nobody, as far as I know," that is worth a conversation.
Reported by:
