00:00Bloomberg has learned that a small group of unauthorized users has accessed Anthropik's new
00:04Methos AI model, which the company itself warns is so powerful it could enable dangerous cyber
00:09attacks. Let's get more on this scoop with our Bloomberg cyber reporter, Maggie Murphy. So what
00:14do we know about this? Yeah, so a source tells us that ever since Anthropik announced that it was
00:22going to release this really powerful model to a select group of users and police it very strictly,
00:30that actually a bunch of AI enthusiasts have had access to it and have been able to play around
00:36with it on a Discord server. Tell us about what they were doing. You say play around, but do we
00:44have to just rely on the kindness of strangers not to do evil things with this supposedly very dangerous
00:50tool? They've been using it to see the capabilities of Mythos. In a demonstration, they tested how
00:59they could generate websites. They say that they haven't used it for bad and used it to look for
01:07any flaws or any of the offensive cyber capabilities that Anthropik was concerned about people being
01:15able to use if they did release it more generally. But it does raise questions that, you know, if this
01:21group of people have been able to get access to it, are there others who have more malicious intent?
01:28I mean, that almost seems like something that you should expect, right? This is technology that's
01:36capable of identifying exploiting vulnerabilities in every major operating system, every major web
01:41browser when directed by a user to do so according to Anthropik itself. Is there anything that they
01:48could be doing more to be making this more secure? When Anthropik told us that they were planning this
01:56release, you know, I had an interview with them and I spoke to the people who were in charge of
02:01making
02:02this decision to not just tone down the model, but keep a version of it that was quite as capable
02:11as
02:11they're claiming and share it with others. And I said to them, how do you know that once you share
02:17it with the 50 or so organizations that you are trusting, do you lose control of it then? And they
02:24said that they have thought very carefully about this, that they had put some measures in place to
02:30ensure that it didn't go wider than the group that they anticipated to be testing it.
02:36But there is a rule in cybersecurity that nothing is ever 100% secure. And so with the, with more
02:45people, there becomes more vectors for, for, you know, people to possibly get access to this. It's
02:52hard to say whether this will go further or not, but I do know for certain that there are malicious
02:57actors out there who are desperate to get their hands on it in much the same way that a lot
03:02of the
03:03organizations that we know and trust around the world want to get their hands on it just so they
03:07can use it for good purposes. This is like the golden ticket at the moment, getting your hands on
03:12Anthropics Mythos model. So there's going to be a lot of very interested parties doing whatever they
03:18can to try and get access.
Comments