I missed the intro of Claude’s Mythos, and only started hearing about them in the ensuing panic. This article has a good rundown of the issues that have the cybersecurity world nervous. I do recognize that this could simply be hyperbole meant to bring extra attention (marketing) to the product, but I thought it worth looking into deeper.
Anthropic just made AI scarier
Update:
The Vox article above is (inconsistently) paywalled. I was able to get back and view it one more time before it locked me out again. So, let me give a quick summary of the concerns here.
“Claude Mythos Preview was designed for general use, Anthropic says, but during testing, the company found it extremely effective at identifying vulnerabilities in the security systems of all types of software, creating potentially massive security concerns.” So, Mythos is the ultimate hacker.
With this in mind, Anthropic has limited Mythos to a handful of tech companies and banks (Project Glasswing), with the hope that those orgs will be able to shore up their defenses before Mythos exposes their vulnerabilities.
“How is Anthropic choosing who to share this technology with?
They’re essentially looking for cyber defenders or companies that a lot of people depend on, and that downstream it would be a huge issue if they got hacked in any way, shape, or form.” It looks like the folks at Anthropic are deciding for themselves, which is all well and good. I think it would be good to slowly roll it out to other orgs, for the same reason. Why is Chase allowed to patch their vulnerabilities while my local credit union is not?
Anyway, these tools are more effective than we think, and in ways we have not imagined. I’m glad that, at least in this instance, we had the ability and wherewithal to be cautious.











