[00:00:20] Nathan Wrigley: Hello there and welcome once again to the WP Builds podcast. You've reached episode number 433 entitled, Feeling insecure with Tim Nash, episode four, the rise of AI in the security space.
It was published on Thursday, the 14th of August, 2025. My name's Nathan Wrigley, and in a few short moments I'll be joined by Tim Nash. But before then, a few bits of housekeeping.
If you like what we're doing at WP Builds, check out our subscribe page. It's wpbuilds.com/subscribe. Over there you'll find all of the places that we publish our content. So it's things like YouTube, it's things like X, Mastodon, Bluesky, that kind of thing.
But also you'll be able to subscribe to our newsletter. We'll send you two a week. One when we produce this the Thursday podcast episode, that's what you're listening to now. But also every Monday we produce a show called This Week in WordPress, I'm joined live by usually three notable people in the WordPress space. And then we package that up on a Tuesday as a podcast episode, so we'll send you an email about that as well.
Speaking of This Week in WordPress, if you fancy joining us for that, we'd love it. This Week in WordPress can be found at wpbuilds.com/live, and it's 2:00 PM UK time every single Monday, barring holidays and things like that. So yeah, wpbuilds.com/live. We'd love you to join in the comments there. It really does drive the show forward and make it more interesting.
If you're listening to this podcast, there's a very, very, very high chance that you are into WordPress. Goodness knows why you're listening to this if you are not. That means that you're a WordPresser, and if you're in the WordPress space and you have a product or service high chances that you'll want to get in front of that audience. We have sponsorship options advertising, if you like. We insert those into the podcast and you're about to hear from three companies that have taken us up on that.
The reason they've done that is because we have a WordPress specific audience. We've built up a bit of trust over the years that we've been doing this podcast, and if you'd like to be a part of that, head to wpbuilds.com/advertise to find out more.
The WP Builds podcast is brought to you today by GoDaddy Pro. GoDaddy Pro, the home of managed WordPress hosting that includes free domain, SSL and 24 7 support. Bundle that with The Hub by GoDaddy Pro to unlock more free benefits to manage multiple sites in one place, invoice clients, and get 30% off new purchases. Find out more at go.me/wpbuilds.
We're also joined this week by Bluehost. Bluehost, redefine your web hosting experience with Bluehost Cloud. Managed WordPress hosting that comes with lightning fast websites, 100% network uptime, and 24 7 priority support. With Bluehost Cloud, the possibilities are outta this world. Experience it today at bluehost.com/cloud.
And also, and finally, we're joined by Omnisend. Omnisend, do you sell your stuff online? Then meet Omnisend. Yes, that Omnisend. The email and SMS tool that helps you make 73 bucks for every dollar spent. The one that's so good, it's almost boring. Hate the excitement of rollercoaster sales? Prefer a steady line going up? Try Omnisend today at Omnisend.com.
And sincere thanks go to GoDaddy Pro, Bluehost and Omnisend for their continuing support of the WP Builds podcast. Podcasts like this cannot happen without the gracious support of companies like that.
Okay. What have we got for you today? Well, at the top of the show, I told you that we were being joined for the fourth time for our Feeling insecure episode. It's Tim Nash, a real expert in the WordPress security space.
And largely today we are talking about AI. To begin with, we talk about WP Security 101, tim's course. And then we dive into AI and all of the ramifications of AI, and how it's making the WP and generally broader security landscape, more muddy.
AI is able to come up with compromises and find vulnerabilities in record time. It also means that the LLM manufacturers, for example, Google with their Gemini, have to figure out how to stop the prompts from giving out information which can compromise systems.
Also, we talk about emails and how now it's trivial to get the grammar and the syntax perfect for people trying to phish your email. That's not good.
And there's a whole load more as well. Always expert advice from Tim Nash, and I hope that you enjoy it.
Hello, there it is the fourth episode of Feeling Insecure. It sums it up perfectly. I think that title, and I'm joined as I have been for the previous three episodes by Tim Nash. Hello, Tim.
[00:05:08] Tim Nash: Hello.
[00:05:09] Nathan Wrigley: Nice to have you with us again. I know that I ask you to do this every time, but there are people who listen to this and, have not listened to our previous endeavors. So just tell us who you are, Tim. And because we're talking about security, explain why people should listen to you about this subject.
[00:05:24] Tim Nash: First of all, if you haven't listened to us before, we have a three other excellent episodes you can go back and listen to. So you, but, my name's Tim Nash. I'm a, WordPress security consultant. I specialize in helping people with code reviews and security reviews. And I also, do training for people.
And I've recently set up a site called Day Security one oh one.com, which is a online course. For all your security wonders, as for why you should listen to me. Oh, you absolutely shouldn't. You should always do your own research. You should always take everything we say with a pinch of salt and go and check this out, because if you don't do that, you are just relying on two blokes on the internet.
However, I have been doing this for many years. I started off before my WordPress endeavors as a penetration tester, and I have, when I started in WordPress, I, worked, Set up a development agency and then decided that was probably too much like hard work. and so went, ah, I'll go and retire for a little bit.
And then went off for a little while to work for a hosting company, managing their. Platform infrastructure and security for their WordPress security platform. And for the last nearly five years, I've been working as a security consultant for companies from really tiny companies through to large enterprises and NGOs that you will have definitely heard of.
And clearly I can't tell you about.
[00:06:51] Nathan Wrigley: Okay. Yeah, yeah, some of that is, is under some sort of non-disclosure agreement. you smuggled in very quickly, and I might say incredibly. British Lee, you, you smuggled in your WP security one oh one.com website. Mentioned it and then quickly moved on. But I'm not going to allow that because I think it's important that we, there's a bit of a quid pro quo here.
You are giving up your time and in exchange I would like to promote what you're doing over there. So the URL is WP Security 1 0 1, so the digits 1 0 1. There's no hyphens or anything like that. And, if you go over there, you're gonna see that Tim has got a course available. It's called WordPress Security Fundamentals.
and is it a kind of thing where you are corralling people through, shepherding people through at the same time in kinda live sessions? Or is it all just consumable? You purchase the course and you can watch it a la carte when you like.
[00:07:43] Tim Nash: You can watch it a la carte as you like. There are a couple of, we've been doing live bits and pieces of, as we've been, getting people through. We're currently releasing, new modules every couple of weeks and new, content every couple of weeks in there. The ultimate goal is that, once you complete the course, there is going to be, there's hands-on labs and there are quizzes and all those things that you expect from like a.
That style of course, 'cause it is meant to enhance your professional development. So if you are at an agency and you are, you've been given professional development training time, this is the course for you. 'cause, you can hopefully use that as counting as credits towards your professional development.
At the very end, we'll give you a shiny certificate and everything that says that you've completed the course.
[00:08:29] Nathan Wrigley: Yeah. So the idea, your ideal customer then is people who obviously have a, an interest in this, maybe tangentially, but want to upskill. So go from not zero, but something. To a lot more over a short space of time. And because it's a la carte, you can dip into it whenever your professional development time allows.
At the moment, the website is advertising a caviar mTOR. I dunno if Tim is gonna amen his prices, but it's, written up in pounds at the moment. You can get the WordPress Security Fundamentals course for 1 9 5, which covers, I'm not gonna tell you what it covers. 'cause if you scroll down you'll see that there are nine modules, which, Which are available to you and over 10 hours of video content, practice labs, and additional resources. So there we
[00:09:13] Tim Nash: to 24. Now,
[00:09:14] Nathan Wrigley: It's closest. Okay. You need to update that number on the website.
[00:09:18] Tim Nash: te tell there's 10, there's about 10 hours of main content, and then
[00:09:22] Nathan Wrigley: Okay. to be fair to you, it does say 10 plus hours. So it's definitely 10 plus hours.
[00:09:28] Tim Nash: I first started I was like so confident about how, I was going to make this a really short course and, 'cause it's just the fundamentals. and so I, proudly put on the sales page before we launched four, four hours of video content.
[00:09:45] Nathan Wrigley: at a, 20.
[00:09:46] Tim Nash: Yeah,
[00:09:47] Nathan Wrigley: Yeah. So that's a full day. So obviously if you are needing a bit of professional development, security's your thing. This could stretch over many, months and weeks and
[00:09:55] Tim Nash: That's the goal is that you are, you can take it at your own pace. And I say we're adding new stuff all the time as well. and the ultimate goal is that, anybody can take this. It is aimed a little bit more, I think it's been branded at WordPress. WordPress professionals, but that's obviously a very broad gamut of people.
That does not necessarily mean you have to be a developer, or a system administrator, though it does cover go quite in depth into some aspects for developers and some aspects for system administrators. But we've got people on there who are work for agencies. We've got people who work for themselves.
we've got one person who just. Bought it and said that they just wanted to do, learn something new, and I was like, oh, that's great. I feel like maybe this might not have been the course for them, but they are absolutely,
[00:10:46] Nathan Wrigley: They're happy. Yeah. Who's, yeah. Yeah. Don't turn them away, Tim. so, the endeavor of this show then is to what? really the enterprise here is Tim and I do these shows, let's say it's quarterly, something like that. We, a, allow about three months or so to elapse, and then Tim joins me and we go over some things.
In the more recent past during that time that Tim has highlighted as interesting, noteworthy. So they, we are really not treading on the toes of the courses where you're trying to build up this corpus of knowledge. This is more exploring what's been happening in the landscape and especially in the WordPress landscape and seeing what interesting anecdotes we can draw out of it.
So where do we begin this time? Time around, Tim?
[00:11:26] Tim Nash: last time we spoke, we said that we were gonna do stuff about ai
[00:11:31] Nathan Wrigley: I
[00:11:31] Tim Nash: and then we got rather distracted as we did.
[00:11:34] Nathan Wrigley: feels somewhat inevitable, doesn't it?
[00:11:37] Tim Nash: Yeah. but, and, as I think what we actually ended up doing was speaking an awful lot about, password hashing and encryption techniques and bits. But, so I thought we'd swing rack round and actually have that discussion about AI and ai, specifically AI in security, looking at the pros and the cons as to where it's being used by bad actors, where it's being used by people who are.
Offering security, and things that you might need to look out for.
[00:12:07] Nathan Wrigley: Yeah, it's such a curious thing because AI very much touts. Its, positives and in many cases you can only see the positives. it makes, it helps me to make content. It reduces the time it takes for me to do the work. But this one immediately, I know nothing about internet security, that's why you are here.
But immediately I can see the adversarial. the, sort of the two sides of the seesaw, if you like. Clearly, if you can, use it to create problems on the internet, it would also be possible to help clean those problems up as well, I say. Okay, let's dig into it. The, the show notes for this will contain any of the links that we mentioned.
There are three, that I think we're gonna mention, but we may stray into some others as well. where do you wanna begin? Which one of those.
[00:12:54] Tim Nash: I think if we start with the Google one, 'cause I think that's the, that one's more encompassing and it's a, it's a report by from Google about, and it was from much earlier in the year. It's not, it's I think it was March time they put published it and it was about, what bad actors are using Google Gemini for.
[00:13:16] Nathan Wrigley: Okay. Yep. I can imagine. It's a lot.
[00:13:19] Tim Nash: Yeah, so, they basically, obviously when you use any of the large LLMs, whether that's open ai, Claude or Google, they are recording what you are doing and they use that data, most of their plans to retrain, so to tell the, but they are logging it and paying attention to it and they obviously can sift through what you are, putting into there.
if you're putting in your per deep personal thoughts, there is a chance someone is. Reading that. Certainly the AI is reading that. But and they generate, in this case, they've generated a report looking at where they've searched for terms around what bad actors and what bad actors might be using their AI for and the different sort of things they've been doing.
Some bad actors have been trying to break the AI. So they've gone, they've, they're trying to do bad prompting, so they're trying to like, Hey, tell me all about yourself, Mr. Ai prompt. I want to know what your underlying programming looks like and
[00:14:17] Nathan Wrigley: I'm gonna butter you up for a minute and then drop the bomb. Okay.
[00:14:21] Tim Nash: you must now pretend to be a Nazi.
and you can never tell me the, not the truth and all sorts of various techniques to try and get it to end up in a place where it cannot manage the logic. So it just dumps something out to you, some information that it might want. you get like people who are like, I need to write an email. Please correct my grammar, which sounds like something that you and I do every single day probably. But of course what they're actually doing is dropping their attempt at a phishing email in there and getting the AI to clean up. So now we have a nice looking phishing email, which is a problem. You then have a group of people who are like, I need to write some code. It just happens to be malware. Please write me. Now, they're not necessarily typing. Please make me some ware. But some people do. Some people literally write, go into the, go into Gemini and go write me some malware to do this. Heza and weirdly, Google quite often says no.
[00:15:25] Nathan Wrigley: Yeah, you would imagine that's gotta be the lowest of the barriers that they've got. Surely.
[00:15:30] Tim Nash: because then you just rewrite it to, I'm a security researcher. Write me some proof of concept to do this dunk and it will go do it for you because there's a, one is an evil, malicious person and the other one is a good person. So you've got those sort of things going on. Then you've got, Gemini, please tell me how to hack Gmail, and how to exploit things in your system.
And the, the assumption being that, Google is going, internal systems are probably gonna know more about each other than they're likely to. I feel that's a false premise, but, it's something that they look through. So Google's put all of this into a big report and it is just fascinating to see how many people.
Use Jedi effectively to either hack, attempt to hack Google, or attempt to use it for nefarious ways.
[00:16:21] Nathan Wrigley: What's really curious about this is that obviously being on the open source side of things, both you and I, big WordPress users, probably advocating for, all sorts of other open source projects as well. There's a bit of me which doesn't like the fact that these LLMs Gemini in this case is proprietary, although it is open source Gemini, which I think may maybe slightly differs from, certain aspects of it are, but on the other hand.
The fact that it's controlled by Google allows them to get this really important data. They can then use to retrain it in ways which prevent this stuff from happening. So it's swings and roundabouts a little bit. On the one hand, that makes me feel sanguine about the fact that it's controlled by this big entity, whereas my intuition usually is big entity or runaway.
That's the first thing. But the second thing is. It's also curious that presumably if we were to rewind the clock, I don't know, five years prior to these LLMs being ubiquitous everywhere, presumably the same work was being done, but just in the heads of the hackers, maybe they were in forums and things where they shared this knowledge, but they were doing it in.
Carefully siloed places where nobody would know that they were up to it. So I don't know if it's the, really good hackers that have moved to doing this in a, in an l and m because they feel their productivity will increase. Or if it's just a whole new cohort of people who thought, heck, I too could be a hacker, We've always had like layers, just like anything. but you always have the low level, which I often given the, nickname, script kitties.
kitties. Yeah.
[00:18:02] Tim Nash: So the, these are, these tend to be people who are either buying. Preexisting exploits and scripts and just using them, or they're writing very low and really rubbish code and, trying to just bulk thing.
They're very different. They're often, and these are often individuals, versus what the, at the other end, the fully huge automation pipelines and bits that can be, actual companies or certainly set up as companies through to actual state actors like. North Korea's packing groups. they are ridiculously sophisticated, very clever, probably still using.
Chat GPT for half their work. But they are at a very different level and probably asking slightly different questions. I suspect the vast majority that we're seeing on the g through Gemini and in this report, especially the ones, maybe not the ones asking about prompts, where it's trying to prompt, break the prompt to try and break Gemini, but the other group, and maybe not the fishing 'cause, why would, you don't need to be secret about those sort of things.
But the ones that are asking about. Make me a proof of concept. I suspect those are less likely to be at that upper level. I suspect they're probably using other tools, whether that is, and remember if you are a group that has access to a large amount of resources, the thing that, the reason that, all of these l LMS are like coalescing around these companies is because they require a vast amount of CPU and GPU.
Processing power to run these very large models at scale. Now, if you are a bad actor who hacks millions of websites for a living, you have a near infinite amount of CPU U usage. Not GPUs, but CPUs. we've only just started seeing websites being hacked for their. CPUs for LLM usage where we're seeing distributed LLMs being put on by malware.
So we, as you're seeing a dip in Bitcoin miners, so Bitcoin miners were where, a bad actor would hack a website and put in a Bitcoin miners. So something that just mines and generates not a, we call 'em Bitcoin miners, but they could be for a range of these sort of. Digital currencies don't necessarily Bitcoin.
They could be, but they basically mine in the background. So they, steal the CPU and the server processes from the site makes your site go very slow. But you might not necessarily know any, anything is wrong. these are normally found by your hosts pretty quick 'cause they can often see a big spike.
we are seeing that those sort of, bitcoin mining has come down partly because of the nature of how all of that has changed over the last few years. But that same idea has now been taken and is being used for LLMs so that the bad actors can have their own large LLM to do. Do whatever nefarious things they want to do, much like the very small one that I have sitting down there.
[00:21:17] Nathan Wrigley: it must be if you were one of the founders of AI back in the day, I suppose on some level. you would be extremely naive if you didn't consider that this would be what a proportion of the people would turn their attention to with the AI that you'd created, the LLMs and what have you.
But equally, it must be fairly demoralizing that you've created this wonderful technology with the capacity to, I don't know, shorten workflow times, cut them in half, cut them by a factor of 10, whatever, create music, create art, and all of that kind of stuff. But a proportion. Of the people are gonna use it for nefarious means, meaning you have to deploy your countermeasures if you like to figure out how the heck to stop them and I presume that Google have, I don't know what the number is, dozens, hundreds, thousands.
I don't know of people who are constantly looking at what Gemini is outputting and trying to Play this game of chess and move ahead of the hackers and figure out what you know, okay, this is the kind of thing they're likely to do. We've gotta stop that. We've gotta make sure that it doesn't respond onto these conditions and so on.
It's just cat and mouse the whole way down.
[00:22:26] Tim Nash: you've just summed up security. It's
[00:22:28] Nathan Wrigley: Yeah. Okay.
[00:22:30] Tim Nash: that is how, that is it. Every time there is an innovation, one side or the other, there, there will be someone who comes along and breaks through that that's the nature of adversarial. It is adversarial in nature and that's how adversarial any adversarial innovation occurs, where you have one group doing something and the other group then goes, I've gotta counter that.
Whether that's an arms race or space race or in this case.
[00:22:56] Nathan Wrigley: Do you have an intuition though, that the, foot has been firmly pushed on the accelerator pedal? Because it feels if you can, do one of the articles that we might mention later sums it up in terms of time, the, amount of time that it took was. I can't remember what the exploit was, but basically a thing that used to take humans roughly in the region of about 17 hours can now be achieved in a series of prompts, which you'll be through that whole thing in five minutes.
So it's not a hundred x, it's not even a thousand x what, whatever that is. It's a lot. and, and so I just wonder. If it gets to the point where you just can't keep up with this stuff. whether the, countermeasures can't be deployed at the rate that humans could deploy against humans.
If it's now AI against ai, whether you can keep up, I don't know.
[00:23:44] Tim Nash: So one of the things that's been a trend even before ai, so was that, as WordPress, and specifically we'll talk about WordPress, but it's, the same generally, but as WordPress has grown, we, it's reached a point where a plugin would, some, there might be an issue in a plugin, causes a vulnerability.
Someone patches that vulnerability. Gets pushed out. Someone announces that they've patched their vulnerability. Now, whether that's somebody like Patch Stack or word fence, whoever says it, or whether it's the person themselves, the time from the person announcing, Hey, there was a vulnerability to someone actively exploiting.
We used to measure that normally in days, if not weeks. So Element, I'm gonna use Elemental elementals, not just because it's a. Plug people will have heard of. But, so let's say there's a vulnerability element, and it gets patched. The announcement goes. 5, 6, 7 years ago that would've been measured in weeks before someone, before a bad actor went, I might go and I'll write something to exploit that and I'll actively start using it.
So over that period, they have to have found that the, information about the exploit, they had to have coded the exploit, tested it, come up with a deployment mechanism to deploy it on large scale, find your site and exploit it. So there was a period of time before, while all of that would be happening now over the years that has come down and I was standing on a stage at Word Camp London in 2019 saying that we measure these things in days now.
So I was saying, oh my God, how scary is it? We've gone from weeks down to days and I was like, if you're doing monthly patching. Bad luck. You are, by the time you've patched, you are really, so you have to be on the ball. We measured these now in hours,
[00:25:52] Nathan Wrigley: gosh.
[00:25:53] Tim Nash: so we've gone from weeks to days, and in the last few years we're down.
We are down to a day, and then we were down into hours. So for a popular plugin. If it has a major exploit that can be exploited, we are seeing those exploits in the wild within hours of the announcements.
[00:26:13] Nathan Wrigley: C, can I ask about that? Have we entirely removed the human from the loop? Because obviously an important part of the story that you just said was the human has to see the thing, has to see the vulnerability. Probably read about it in a blog post somewhere. maybe they've subscribed to word fence or patch deck or whatever it may be on their RSS.
Reader of choice, but they, found it. A few days have elapsed, they've cogitated on it, thought, yeah, that's worth exploiting. Then finally sat down at the computer and so on and so forth. Do we now even need a human in that cycle? Or can a, can an ai for example, be scraping the, those same websites, discover that there's a thing, go and write code to exploit that thing.
Put the thing out there. So the person who would traditionally have been the hacker all the while is just fast asleep. It's four in the morning.
[00:26:57] Tim Nash: We, we, don't even need to use AI for that, that, ultimately you are just scraping sites and pulling out text data. what actually, for a lot of the, there are a lot of, attempts where they're just reading the repos and looking for okay. Yep.
So there is a near constant scanning going on of looking for commits, and there are tools to compare, commits to see what that commit might have done.
So they are looking for vari variations where they're seeing a commit, which may have, might have something like escape, HTML in there as a function. And then we like, huh, why are you adding that to that? because you can see the diffs between them and go, oh, If that wasn't there before, that means the previous version's vulnerable.
So if you are, a bad actor and you're really lucky and you've, you are code scanning across large popular blockage, you might actually know the vulnerability before it's publicly even announced because you've seen the commit that fixes it, and you can just invert one step to work out backwards.
There is still, there are still human elements at play for most of these, but there are plenty of them that are trying automated. The other thing is that if you're doing it at scale, it doesn't really matter if you get it wrong most of the time. 'cause what's the worst that's gonna happen is you don't succeed and it just fails after a few attempts and you go, okay, cancel that and throw that code away.
[00:28:19] Nathan Wrigley: Okay. I guess that's the difference. Now, the scale, you, can just create these patches without any human cost. in terms of hours sitting in front of a computer, you can just come up with, okay, that didn't work. Why didn't it work? Let's just try something else. Okay, deploy that. Let's try something else.
Deploy that. Oh, that now works. Okay. Fascinating. It's okay. So Tim likes to use the Monica Doom speaker. this definitely qualifies under the category of Doom speaking, I think. But also, I suppose there's the other flip side. Google are clearly doing something. About it. I don't know if there's any, any legal compulsion for these ais to do this kind of work, to carry out these kind of things.
Or if they can just push this stuff out into the wild and not do this. Is it
[00:29:06] Tim Nash: think there's a lot of gray and, for the bigger companies, they are tending to be on the safer side because from their point of view, what they don't want is necessarily, they'll publicly be saying, oh, we definitely want legislation over this. We think it's really important that everybody stay nice and safe because these are so super powerful and scary.
Which in themselves is like digging up their own sales
[00:29:30] Nathan Wrigley: Yes.
[00:29:31] Tim Nash: So they want you to think these are huge and big and scary, and not just basically a statistical modeling solution. A lot of LLM power is not that powerful really. and it has existed for a long time. We are just packaging them up nicely now and understanding how we can do this a distributed scale, which is one of the reasons that it's so CPU and GPU intensive, but. They do not necessarily want legislation. They do not want to be forced into, a, into boxes that they then can't break out of. So it's in their interest to control and to be seen to be policing themselves. 'cause if they're not seen to be policing themselves, you already have, like the EU is already talking about how it's gonna legislate for these AI models and.
If we know anything, and I, we know that when it comes to politicians and technology, they are not very good at this.
[00:30:30] Nathan Wrigley: Yes.
[00:30:31] Tim Nash: do not understand technology. So we could end up in a really bad place where some politicians try to do to come up with some laws, and it will inevitably be a disaster. So I think the big companies at least partly recognize for their bottom line, not for your safety, but for their bottom
[00:30:49] Nathan Wrigley: Because I, the they're of trying to avoid the big pr.
[00:30:51] Tim Nash: should be policing this, which is why you do see it.
[00:30:54] Nathan Wrigley: I don't know that we've yet had the, that. where the big announcement has come along yet? I certainly haven't seen it out in the world, on your domestic news channel like the BBC or whatever you read, I haven't yet seen the AI generated problem that's destroyed the internet kind of thing yet.
and I guess these companies, the PR consequences of something like that, something that was. Categorically made by their AI deployed. It took, I don't know, took all email offline or took CloudFlare out or whatever it may be. the PR consequences of that would be pretty profound. yeah, I'm sure it's not, it's not out of the realms of possibility, let's put it that way.
[00:31:42] Tim Nash: it's not. And there is, there, we aren't going to see at least one or two catastrophic things devoted to ai, while it may not be in the mainstream press, there are still plenty of, I let AI buy code base. I don't have a code base.
[00:31:58] Nathan Wrigley: yeah. Okay.
[00:31:58] Tim Nash: oh, you poor. Oh wait, that's like a major company that's just said
that. Yeah, Let's hope it's something online and not something, I don't know, to do with gene research or something
Yeah, there was one company that allowed their AI to, access Git, which is a version control system, which is a great idea 'cause you want your code versions and the AI was automatically versioning things. I think I know the story. It's
then it just decided it would just destroy the entire repo by squashing all the commits into one and getting rid of the history because it was taking up too much context.
'cause ais need information, but if you give them too much information, then they don't have enough space in their buffer effectively,
it just squashed all the commits.
[00:32:43] Nathan Wrigley: I read this story. I think it's probably the same story, and, it, felt like the guy who discovered this kind of woke up the next morning, sat at the computer to discover it happened, and then realized what had happened then gets into conversation with the LLM and the LLM apologized, Yes, I did in fact delete everything. I'm sorry about that. Okay. Oh, At least you're apologetic. It's fine. Okay. So there's the first story, the, the one about, generative AI and the adversarial misuse of it. Where are we going next?
[00:33:20] Tim Nash: so I thought for the next one we would, one of the things that came out in that, Google report was, that. Bad actors are using Gemini and open AI to write their phishing emails for them. And so the next one I wanted to highlight was a, an article on mail gun, which is a email, provider.
So they, they, will send email on your behalf, talking about how phishing emails have changed under ai.
[00:33:52] Nathan Wrigley: I'm guessing in the past we were all, there were certain cues that we needed to receive in an email in order to be alerted to it. that the whole kind of I, don't know. I'm a king in some part of the world, and I've got you $3 million. It's just waiting for you to, at. And obviously your spidey sense is immediately hooked by that, and you think that's implausible, but then it got more sophisticated over the years, that kind of stuff dropped away.
And, but where's the state of it now? What, are the, things, what, kind of level of mindfulness do we need to have Reading email.
[00:34:24] Tim Nash: it's pretty terrifying now, because again, going back to that 2019 talk I was giving, I actually had a pic. I brought up a, an email that, had you need to update your WordPress database. Now I. It's actually not something that we need to do anymore. That's not, but for anybody who used to be in WordPress, we used to get an email when your site updated that said you needed to update your WordPress database and you'd click the button in the email and it would do the go through that update your database. Now that email back in 2019 had a few little telltale signs that something wasn't right. it WordPress was spelt wrong, in three different ways.
[00:35:06] Nathan Wrigley: It's interesting.
[00:35:08] Tim Nash: like they couldn't, they just couldn't cope with the idea of a capital W and a capital P in it.
[00:35:12] Nathan Wrigley: Okay.
[00:35:13] Tim Nash: and at one point they also couldn't remember whether we were on wordpress.com or wordpress.org, which is a problem many of us have.
[00:35:19] Nathan Wrigley: Yeah. Yeah. Okay.
[00:35:21] Tim Nash: So there were little telltale signs in the email.
[00:35:24] Nathan Wrigley: Subtle for those who know, but probably not visible to those who
[00:35:28] Tim Nash: Exactly. And there were grammar issues and there were spelling mistakes and it was like if you were paying attention, this was an obvious email. Now, I was doing some training for a company the other week and I put up a. Email that I'd seen really recently, which is a similar version of the same scam, which was that site is your, site is currently in maintenance mode.
You need, after update, you need to reactivate as one of the plugins has failed. And I suspect you haven't seen that email in WordPress call because that word email doesn't exist. it sounded quite believable, and when you read it, it looked like a WordPress core email.
It had all the same grammar, it had the correct, it had the correct footer and header. Everything about it read well. The giveaway was indeed the magically M dash. Just one, one random M dash. an m dash is the double long, the long, hyphen. I dunno how, if that's the correct
[00:36:38] Nathan Wrigley: Yeah, it's like a, it's like an extra wide hyphen, isn't it? it's the piece of punctuation that almost nobody actually uses. In fact, I don't even know how to get it on my keyboard. If you ask me to make one, I do not know how, but AI seems to love it.
[00:36:52] Tim Nash: Yeah, specifically OpenAI has this thing about it. So that was the only hint in the text that something was wrong. And that was only because, wait, when could this have been put in and that, so this is someone, me who's technical going, this it, this sounds almost logical. This feels like part of the fate.
'cause we, about. Four or five years ago, there was added in a lot more rollback features and failures for updates. So it's okay, this is believable that I've just never come across this email. It, sits there. Oh, no. Hang on a minute. There's that m there. I don't believe that anybody used that before last year.
[00:37:31] Nathan Wrigley: Yes.
[00:37:31] Tim Nash: think that fig that existed as a fig. and that was the thing that, let me realize, hold on a minute. This doesn't look right
[00:37:39] Nathan Wrigley: Okay,
[00:37:40] Tim Nash: you get there, you once the second you see something like that, you can spot all the other red flags.
[00:37:45] Nathan Wrigley: but in your case, so you know what you're doing. That was it. It was the difference between an M dash and not an M dash. So if they deployed that email and somehow figured out, let's not use the M dash and honestly, that's gonna be in the next prompt, isn't it? wherever you, you want to use an M dash, put a comma in a space or something like that.
Okay. That's pretty profound. So we're on the level of grammar.
[00:38:09] Tim Nash: Yeah.
[00:38:10] Nathan Wrigley: We're trying to, so the, whole thing about, I don't know, misused words, incorrect sentence length, just clumsy English and clumsy grammar, that's now fixable in a heartbeat, right? You, put in your clumsy grammar sentence into an AI and it spits out perfect grammar apart from the M dash, but we're approximating perfect grammar, so you
[00:38:32] Tim Nash: you click the link, obviously it's not taking you to your site, it's taking you to their, phishing site. And if you looked at the URL, but bear in mind that a lot of, browsers now hide an awful lot of that URL.
[00:38:46] Nathan Wrigley: Yeah,
[00:38:47] Tim Nash: So it's very difficult to spot. and I can't remember what the URL was, it was something like WordPress, WordPress admin.io or something.
[00:38:59] Nathan Wrigley: pretty good. Pretty
[00:39:00] Tim Nash: yeah, something, enough that you might, if you weren't paying attention, think it looks reasonable as a URL. And the screen that came up was a WordPress login screen.
[00:39:11] Nathan Wrigley: Yeah.
[00:39:11] Tim Nash: Just without like just a couple of bits moved from it, but it looked like the WordPress login screen again a few years ago.
That would've looked awful.
[00:39:22] Nathan Wrigley: Yeah. Clumsy design. Just quickly thrown together, the, it would've shouted from you right Whereas now that's not, so all of a sudden you've got really, you are suddenly like, okay, how was the delivery mechanism? Is that, is it coming from an email that is valid and real? But phishing emails can spoof email addresses. And this, particular one, when I looked at the email address, when you go and look at the email address properly by going and inspecting the headers, and you're like, you, can see that it is from a different email, but the spoofed version said it was coming from my account.
[00:40:01] Tim Nash: On my website, and I can only assume that the reason, I'm not assuming I know how they got, that they validated the email address by, a, my, my own email address is publicly available at some point on my site, so they've grabbed that. They have validated that against a user. So they have gone to, the something like do PJSO, and then gone to the user's API because I have, I'm an author on there and they will have found the Gravita link. They'll have check the email address against that gravita and now they have a user. So when it came to me, it did say, hi T Nash, my username you need to do, So it ha so it's closer to a spear phishing attack than it was a phishing one, even though it's completely a hundred percent automated.
[00:40:52] Nathan Wrigley: So this feels like a profound change. And again, the human in the loop is the, amount of time to create this for, a human is considerable. you'd have to sit down and write an email and what have you, all of this is now automated. The grammar is checked. Where does this even leave us?
So are we now at the point where we basically have to distrust. A, huge proportion of email that we don't actually know the recipient of,
do we have guardrails to make email safe?
[00:41:22] Tim Nash: you, you, should have been doing that anyway.
[00:41:24] Nathan Wrigley: That's true, but there's, there, there's always a point. So what I do, I, use Gmail as my client. I always, if, I'm, if something just doesn't give me that confidence, like it's basically it's somebody that I've received email from before or a service that I know I've subscribed to, I will go and click the little gray down arrow and I'll expect, I'll inspect the domain that it's coming from.
I don't know if that. Makes me any more secure, but I, will check it. Letter for letter. So let's say it's Google. I will check that it is something.google.com, that it's perfectly, that without any kind of weird thing, the WordPress hyphen admin or what have you. But I don't think, I don't think anybody else that I know is doing that.
Maybe that saved me once or twice. I'm relying on my email client. Again, it's Gmail if, or Google's workspace, relying on that to sift out most of the dross. But, I'm still, I'm, I still think I am probably more safe than most because I go to those lengths, but I don't think my, I don't think the majority of my family will be doing that.
And so it really does that, bulwark of, oh, this looks weird. It sounds weird. The grammar's wrong. Okay, this weird been it. That feels like those days are, if not disappearing, very soon to disappear.
[00:42:42] Tim Nash: Yeah.
[00:42:44] Nathan Wrigley: Okay.
[00:42:44] Tim Nash: And unfortunately at the moment, the email, providers are at the disadvantage. So you may have noticed that the amount of spam getting through in just spam, let alone phishing attacks, getting through, things like Gmail has. Se seemingly increased, as they are under exceeding more pressure and things are getting more sophisticated.
There is a really good rule that if you weren't expecting an email with a link to click on it, don't click on that
[00:43:13] Nathan Wrigley: Yeah.
[00:43:14] Tim Nash: That I, that is the golden rule. It doesn't matter if you get an email and you weren't expecting it, and it's got a link on it. Don't click that link, especially when it comes to WordPress, because whatever it's asking you to do, if you just go into your website, it will prompt you so you are not there.
There is almost no time in the, when you are doing anything to do with WordPress, where you have to click that link.
[00:43:37] Nathan Wrigley: Yeah, that, that's the kind of message that I've tried to inculcate in my family members is basically, if you get an email. Let's say from your bank or something purporting to be your bank. It's no hardship to go to your bank website. Just open the, open a new tab and type in bank.com, whatever that may be, and then just do that thing yourself.
if, there's a actual notification, it'll be in there somewhere if it's there, but basically don't trust it. But obviously now that the whole, the ability to pause the grammar and kind of think, oh, that all just sounds a bit fishy, that's fast disappearing. So presumably some of that's coming through.
I got an email the other day that caught me off guard, and it was interesting because they couched it in language, which, and it was definitely. Spam and possibly some sort of phishing, but it couched it in language, which kind of purported, it made me feel a bit secure. At the same time it said, we will never ask for your username and password.
And I thought, oh, that sounds good. And then I noticed there was a link beneath it and you think, okay, you're kinda giving with one hand and taken with the other. So you've said, we will never ask for your username and password, but just click this link, then give us your username and password. But it, on some level, it worked.
Some part of my brain got. Hijacked by that, and I thought, oh, okay. Wait, It's amazing how clever they are.
[00:44:56] Tim Nash: I guess psychology of a bad actor and when they are good. That's exactly you, and we almost get lulled into a false sense of security by all the rubbish that we see. but the reality is that there are people who do study this, who, and they treat it like a bit. If you imagine, if you are a company and you are marketing, doing marketing analysis and all that business stuff, ban actors do that too.
[00:45:25] Nathan Wrigley: Yeah.
[00:45:25] Tim Nash: For them. It is a business and I think a lot of people. Think of these as like a person in a basement going, he, I'm going to cause
some it's not like that. is, It
is large and a very profitable business, for them. So they will treat it like a business. They will refine their truck, their craft, they will refine or they will be getting feedback on all of this.
They will, and they can. Regurgitate and push it through. And now we have these lms, which are doing all the things that help you empower your businesses, helping empower their business as well.
[00:46:00] Nathan Wrigley: they only need to get you once as well. That's the other thing. they can spray out a million emails and what only need to get one of those or none. they try, tomorrow and get one of them instead. that's the brutal reality of it as well, isn't it? Just click on one thing at one point this year and you are, you could be hosed.
Oh, it's depressing. My advice at this point in the podcast is just turn the computers off, switch 'em all off. Unplug. Don't, do anything
[00:46:29] Tim Nash: this is where if we were on, Nathan's, weekly show,
[00:46:35] Nathan Wrigley: this
[00:46:36] Tim Nash: he would be, showing, pictures of kittens and bunny rabbits.
[00:46:40] Nathan Wrigley: right. Yeah. we are trying to, calm you all down. So everybody pause, go and look at some pictures of kittens. Come back 'cause we've got one more. We have one more link. What's the last one we're co covering today? Patch Stack, I believe.
[00:46:54] Tim Nash: Yes, so this is, patch I, I both patch stack and word fence. And I think actually there's a couple of others now do a regular security review. And so this link is actually not from them directly, it's a article in the repository.
[00:47:12] Nathan Wrigley: Oh, that's right. Yeah.
Thank
[00:47:13] Tim Nash: and it's basically just a summary of their, their review now. Patch stack are quite good.
They have very pretty graphs and they make everything look pretty when they do their reviews. I, they are the sort of people that I quite often will refer, show clients their graphs and things, and they're all nicely attributed. So I can just copy 'em and go, look, there's a pretty graph that's.
Should terrify you. And this is pretty much the same thing where we're, we are looking at a review that says that the number of, vulnerabilities as report has sharply risen. they have a little bit of a blame of AI amongst that.
[00:47:51] Nathan Wrigley: Okay.
[00:47:52] Tim Nash: and that the levels and the scariness of these have also slightly risen.
[00:47:57] Nathan Wrigley: And, presumably this is gonna be a thread running through all these stories that AI is partially responsible for. This is, can you account for it in any other way? if it, is it purely that AI is being injected into WordPress or is there some other factor that may be involved?
[00:48:14] Tim Nash: So there is certainly a potential that AI is helping to, create proof of concepts, quick, more quickly, that the AI itself, is therefore reducing the amount of time it takes to do the exploit. But these are also, we are seeing, increases in. larger company, larger hacking groups, taking, exploiting.
And when they do they often are using, utilizing existing frameworks, existing automation tools, and already have access to your site. So we are seeing that sort of increase in part because of the AI being quicker actually writing things. But that there is now a much, stronger and better set up framework for deploying these exploits.
So we are seeing that it, the time it takes, they're not having to rewrite, recreate the wheel every time. They can just throw this up onto their normal exploiting system. We are also in a cascading event where, more websites that get hacked, the more powerful everything becomes, the more easier it is for them to deploy on a larger scale.
So it, all cascades and gets faster and faster as well. But the, time aspect that possibly is either down to AI or that there are now a large amount of unemployed devs who have taken up hacking WordPress
[00:49:32] Nathan Wrigley: Okay. Sanguine as always. That's great. So who's. What do we do about all this? So obviously we take your course and we figure out some of the kernels of knowledge that we need to do this, to protect ourselves, who understand it at least. I noticed that at the bottom of the article, Ray who writes the repository, articles, makes the point that it would be nice if.
Hosting companies played a more proactive role in that. I don't know what you think about that, whether that's just offsetting the blame. It's open source after all. caveat mTOR, you download WordPress, you're on your own. You used to work for a hosting company in this role. What are your thoughts on that?
Should hosting companies take, the blame when it all goes
[00:50:16] Tim Nash: they certainly shouldn't take the blame. the thing is, I think most people don't understand that hosting companies don't want you to be hacked. Because it costs them money. They don't want their resources to be used. They certainly don't want, they, they don't really mind if you are suffering from ransomware or similar, there are certain attacks that, that they, have less interest in, but the second that their resources are being abused, they want to get that fixed quite quickly.
generally hosts. We'll help you. They'll help clean up your site. They could, many of them could do more with preventative measures, but everything with security, it feels like a set of compromises. If hosting companies say, we will automatically update you, which is what the hosting company I worked for, we mandated you will be updated.
We'll give you up to 14 days to opt out of it, but you will be updated to the latest version of WordPress. You will be update, your plugins will automatically be updated. We'll give you a staging site, we'll give you testing mechanisms, but you will do this. and consequently, we had far less problems. We also had people leave the, because they were like.
We can't work in this environment. We need stability. We can't have this, we have to update once a month. And in fact, one of the cases, we can't do this for compliance reasons because our, we can't get our testing done in the period of time. It's this is where a balancing act comes forward. Likewise. A little like oddly dirty secret.
The cheaper the host, the more aggressive their waf, their web application firewall's gonna be, the more likely it is to block things.
[00:51:59] Nathan Wrigley: Interesting.
[00:52:00] Tim Nash: Because there you are gonna be thrown on a shared host. So that's a, it's a server that has lots of hosting hosts, lots of websites hosted on it. and they all share resource, the same set of resources, and they're not containerized in any way.
So they want to, if, a one of those sites gets hacked and there's. a serious possibility that all of them will get hacked through this approach. So they do want to try and avoid that. And the way to avoid doing that is to, put really strict fire firewall rules in, but it does mean that when you try to run that weird little plugin.
[00:52:39] Nathan Wrigley: Yeah.
[00:52:39] Tim Nash: triggers the firewall rules and you end up in this scenario, we have to contact support and say, I can't use my export function of my plugin because it says it's triggering this rule you get more expensive hosting. The WAFF rules get less and less because they want to give you, the customer what you want and it's not costing.
And because you are not going to have resources being taken from other users, it's okay. You are nicely isolated and there's the more isolated you become, the less the host is worried about you. 'cause the worst case scenario is they can turn you off.
not gonna affect the resources of everybody else.
So you do end up in this really weird scenario where people who pay the most often have the least security.
[00:53:21] Nathan Wrigley: Isn't that's totally counterintuitive in my head, but I get it. You've explained it well.
[00:53:27] Tim Nash: so can hosting companies do more? Absolutely. Should they be responsible for your security? No, it's everyone's responsibility. They do have a role to play, but so do you. And honestly, the answer to what can we do is exactly the same as it was. Six years ago, 10 years ago, and that's keep everything up to date.
I, without trying to, sell you a course, which will tell you the same thing, you keeping everything up to date. And if you can switch to automatic updates, obviously there's lots of controversy over whether these are the right approaches and they, not for everybody, but for the vast majority of people turning on automatic updates. Will work for you. There are tools like Patch Stack, like Wordfence that provide some effectively virtual patching, but these are partial mitigations and, you, they are less effective if you've got automatic updates on. They are very, they're far less effective. If you're just leaving the site going, then something like that might help you.
[00:54:34] Nathan Wrigley: The, one of the things that Tim and I chatted about prior to hitting record was the, but we won't be able to get into it today. I was. About sat Tim, was the, whole scoring mechanism for how, vulnerabilities are scored. and I know the very merest amount about that, but I'm thinking maybe Tim, can we put a pin in that topic and come back to that?
Maybe that could be the, topic of, a, at least a bit of next time. That because that was Tim revealed something quite interesting to me about how it scores are maybe self-inflating. And not, maybe not as severe as they as you think they are. So should we do that one next time? How does that sound?
[00:55:13] Tim Nash: That sounds good. I like how that's nice. Nicely leaving somebody on tender hooks, Yeah. Oh yeah. Okay. You've only got three months to wait. Yeah. was there anything that we missed there? did I miss anything in that last article or do you think we've
no, I think we've covered it. I think we, we had like huge chunks of stuff to talk about ai and it's one of these things that we could go on for. One hour, two hours, 10 hours and still not have covered the gamut of things because everything is changing so fast. And like in three months time, we're probably gonna be saying, and this crazy thing has
[00:55:47] Nathan Wrigley: Yeah. Yeah. It, never seems to slow down. I am, I listen to a security podcast and, they're, well over a thousand episodes now, and right at the beginning they thought it would be maximum 15 minutes each week, and now they're approaching the two and three hours. Each week with every bit of content is, something brand new.
They do a bit of a deep dive, but the point is it never slows. It only gets, more crazy. we'll be back, I guess in about three months time for episode five of feeling insecure. It only remains for me to say, firstly, go and check out, Tim's course. WP Security 1 0 1 numerals.com.
There's no weird. M dash is in that URL, just WP Security one oh one.com. Go
[00:56:40] Tim Nash: know if an M dash would be valid in a
[00:56:42] Nathan Wrigley: I don't even know if that's, is that, admissible in a UR? I Probably. I've never seen one put it that way. go and check that out. But also just from me, Tim, thank you so much for joining me and enlightening us, scaring us, whatever the right word is.
Thank you very much. Cheers.
[00:56:59] Tim Nash: a lot of fun.
[00:57:00] Nathan Wrigley: Okay, that's all I've got for you today. I hope that you enjoyed that. If you did enjoy it, head to wpbuilds.com. Search for episode number 433, and leave us a comment there. As I always say, if you don't like it, always go to that URL as well and leave us a comment. You know, we'd love to hear from you whether you like it or not. Be interesting to hear what you've got to say. Use our WordPress comments, that would be the best way to get in touch.
But don't forget wpbuilds.com/subscribe. There are other places where we'll post that content as well, and we will no doubt contribute to the conversation that you start over there as well.
The WP Builds podcast is brought to you today by GoDaddy Pro. GoDaddy Pro, the home of managed WordPress hosting that includes free domain, SSL and 24 7 support. Bundle that with The Hub by GoDaddy Pro to unlock more free benefits to manage multiple sites in one place, invoice clients, and get 30% off new purchases. Find out more at go.me/wpbuilds.
We're also joined this week by Bluehost. Bluehost, redefine your web hosting experience with Bluehost Cloud. Managed WordPress hosting that comes with lightning fast websites, 100% network uptime, and 24 7 priority support. With Bluehost Cloud, the possibilities are outta this world. Experience it today at bluehost.com/cloud.
And also, and finally, we're joined by Omnisend. Omnisend, do you sell your stuff online? Then meet Omnisend. Yes, that Omnisend. The email and SMS tool that helps you make 73 bucks for every dollar spent. The one that's so good, it's almost boring. Hate the excitement of rollercoaster sales? Prefer a steady line going up? Try Omnisend today at Omnisend.com.
And sincere thanks go to GoDaddy Pro, Bluehost and Omnisend for their continuing support of the WP Builds podcast.
Okay, that's nearly all I've got time for. Just a quick reminder. We'll be back next Thursday for an episode of the podcast. It'll be a guest interview.
We'll also have our This Week in WordPress show that will be on Monday, UK time, wpbuilds.com/live.
If you've enjoyed this and you'd like to advertise and sponsor the podcast ../advertise.
And that truly is very, very, very nearly it. A big thanks to Tim Nash. I'm gonna fade in some cheesy music and say, stay safe. Have a good week. Bye-bye for now.
