John Artunkal 0:00

People like authentic candidates. They like people who are a little rough around the edges. The only thing is you've got to make sure that you don't have any career enders.

Eric Wilson 0:11

Welcome to the Campaign Trend Podcast, where you're joining in on a conversation with the entrepreneurs, operatives, and experts who make professional politics happen. I'm your host, Eric Wilson, and my guest today is John Artikel, founder and CEO of Argus AI. John came into politics through kind of a side door. He was trained in the open source uh intelligence field by Bellingcat. It's an outfit sort of better known for identifying Russian intelligence officers from selfies and flight manifests. And he did his master's in the social science of the internet at Oxford University. Now he's applying that tradecraft to candidates and campaigns. On today's episode, we're looking at how AI is and isn't transforming opposition research in politics. We're digging into what more affordable Oppo does to candidate recruitment. Why self-fetting is now table stakes for down ballot campaigns, and what the open source world knows about politics that hasn't caught up yet. And whether you could chest a machine to tell you your candidate is clean. So, John, you were trained in open source intelligence, some people call it OSINT or OSINT, um, by Bellingcat. It's sort of a different lineage than the traditional mix of you know looking at legislative voting records and C-SPAN tapes. What what does that methodology look like when the subject is, say, a state senate candidate instead of some sort of foreign military actor?

John Artunkal 1:44

Yeah, absolutely. So the firstly, thank you very much for for having me on. And um, there's a lot of the same tools that you would use regardless of who they are. So, you know, one classic tool people use to either verify who someone is or work out what associations they have is facial recognition tools. And one thing I noticed uh last year when I was uh a contractor for a campaign, a gubernatorial race in the US, one thing I noticed was that these facial recognition tools are being underutilized. And actually, they're a really good way of finding things like trackers, uh, seeing if someone's a paid or professional disruptor, but also for a particular candidate, it's a very good thing to use to see what associations they had in the past, and indeed if there's anything uh compromising. Another big aspect of uh OSINT is geolocation as well, which is trying to work out where a photograph uh was taken, um, to see if that has any information that's particularly uh relevant too. It can also help you identify someone, what is in the background of the image that tells you where they are in this particular moment. So those things are quite useful.

Eric Wilson 2:53

And and I think it's important that we we, you know, because you're talking about some kind of spy stuff, but we should be really clear that open source intelligence means that it's you know, someone's posted it, it's publicly available, you're not uh hacking into systems, you don't have like a Palantir database from the government. Uh you're you're looking at all the things that people post publicly, right?

John Artunkal 3:17

Yeah, absolutely. I think um there's a skill to learning how to use these tools and where to source the data from, whether it's images or emails, mobile numbers, but these are tools that are at all of our fingertips. It's just a matter of knowing where to find the information and how to then seek more of it.

Eric Wilson 3:36

And so in in the political world, we kind of have our own established playbook. Obviously, you come from a separate discipline. What what what are the things that are sort of table stakes in OSINT that that really haven't made its way into politics yet? You mentioned facial recognition, geolocation. Are there other things that that that are known within that toolbox that we we aren't taking advantage of?

John Artunkal 4:02

Yes, absolutely. And the biggest one, Eric, is digital footprint mapping. So, you know, traditional Oppo research and vetting could involve, you know, hours of conversations with other human beings. It could be looking at, you know, FOIA requests, looking at legal records, or as you were mentioned, someone's voting history. But really, if we think about politics in the last few years, some of the biggest uh Oppo research um stories have been about someone's digital footprint. And what that means is that someone like me has built a seed list of someone's email addresses, the usernames they use, the mobile numbers they use, IP addresses, even passwords, and has tried to build out a complete map of that person's online presence where they have accounts. Do they have accounts in an online forum or even a dating site where perhaps they they shouldn't have them? It's these sorts of things that sometimes have the biggest splash when it comes to to opera research.

Eric Wilson 5:01

It seems to me, based on what I know about OSENT and and increasingly digital footprint mapping, is uh it's like a thread or or breadcrumbs where you sort of start following it. And so, for example, just to give our our our listeners a an example, someone uses their username over and over again. And so maybe they think on this side, oh well I'm anonymous, but then on another site they use the exact same username tied to their real name, their real photo, and you you you search that and now you you find the next stone along the path.

John Artunkal 5:41

Yeah, and uh I you know, I think we were discussing previously a great example of that was the the Mark Robinson case in in 2024, where he not only used the same username, uh, but he also gave more leads as well. So he gave a lot of biographical information, um, specifying how many years he was married, um, his hometown, his mother's occupation. So these things can all be cross-referenced too. The other thing that is sometimes very useful is also linguistic forensics. If someone has a kind of phrase that no one else really uses, or they make the same misspellings that nobody else uses, you'd be really surprised how often people are found out through those methods as well.

Eric Wilson 6:25

So I I want to turn our our focus now to bringing AI to bear in in this uh endeavor. And and it's kind of a mirror of what we've seen in other places. We've talked previously on the show about AI polling, for example. Um and and and there's there's kind of a pattern that's emerging. So so campaign right now in the world of research has has kind of three options. You know, there's no research at all, we can't afford it, it's not in our skill set, we don't have budget. Or the sort of gold standard human vet that's gonna cost thousands of dollars. Most campaigns can't afford that. But then there's the AI output that's fast and affordable, but yes, imperfect right now. Um that middle tier is getting really uh more common for campaigns. And and so I'm curious, as someone who who is filling that that space, how should a campaign think about what they're they're actually buying, the trade-offs, what they're giving up at those those different levels?

John Artunkal 7:30

Yeah, absolutely. So um the first thing I would say is if you're a campaign with basically zero budget, um, you really can't afford anything, you can't uh afford any human intelligence or any open source intelligence, your best bet is to try to find uh a sort of uh a bright, scrappy online, uh chronically online college students. Um, because there are you'd be surprised how many people um are just natural open source intelligence investigators. Um, you know, people who just spend enough time on social media uh have quite good instincts uh on finding people's accounts. So that would be like your first layer. And ideally, what you would want to do is first have that person set their sights on the candidate themselves. So I think self-vetting is is really critical. You need to know if there's something that could utterly destroy the campaign. Um but um beyond that, if you do have a bit of budget and you want to go for the open source intelligence route, the digital footprint route, there is actually quite a lot of range uh within that market as well. And ideally, you want to go for tools that are designed for politics, for opposition researchers, um, and that are a little bit more bespoke than just simply you know the kinds of tools that you know uh an HR department might use when vetting in an employee. So that's sort of the the middle tier. And again, you can first bring it into the campaign and then you can move it on to the um opposing candidate and to their team members as well. And then finally, you know, the the more most time-intensive one, and therefore in some ways the most costly, is this human intelligence layer. Um, and while there's so many great findings that you can have using open source intelligence and digital footprint mapping and looking at legal databases, there are some wins that only a human source can give you. And that's sort of the the tops here at the end.

Eric Wilson 9:23

Yeah. I so I I think the lesson is fine, you know, young people are particularly adept at stocking potential love interests online uh these days, and so you can bring those those skills to bear for your campaign. One of the things, John, that is really fascinating to me uh as someone who thinks a lot about product and and software in in politics is that the OSINT world has these sort of open source, freely available tools that they kind of collaborate on, and that's the that's just not something that we're we're as familiar with. Can you can tell us a little bit about you know, like what are some of the problems uh that that the community tries to solve and the the tools that have been built for that?

John Artunkal 10:05

Yeah, absolutely. I think in part the the OSINTS community was built on trying to uncover things like you know sanctions busting or human rights violations. And so obviously it's a lot easier for them to come together and say, you know, some of us are professional practitioners, but a lot of us are journalists and activists, and we're trying to solve these problems together. We're gonna share our tools, and obviously some people have monetized them very successfully as well. In the political sphere, there's been less work on that, and that's partially because politics is by nature completely divided. So, you don't, you know, if you're a Republican, you've built a great tool, you may not necessarily want Democrats to be using it and and and vice versa. Um so that's definitely uh an issue, but I will say that I think um it is in the interest of everyone in politics that candidates are at least of a certain caliber, and we all have a public interest of making sure that they they surpass that certain caliber. We may have disagreements on the agenda. Uh so I think it would be good for those on the political sphere to share some of their skills and tools together as well.

Eric Wilson 11:08

You're listening to the campaign trend podcast. I'm speaking with John Artinkle about the future of AI in opposition research. So, John, I think as we expand these tools, uh we are we are starting to move further down the ballot. So most candidates right now below the federal congressional level have have never been vetted by anyone, not their party, not a firm, not themselves. What what's actually sitting in those digital uh footprints that that people should be aware of?

John Artunkal 11:42

Well, you know, sometimes people are are boring or you know, they they they they're clean, and you know, there's sometimes there is you know nothing there, and it's kind of what you hope for. Uh but unfortunately, um, and this is certainly something we've seen in the UK in recent local elections, you also have things like you know, white supremacist dating apps, or you know, uh sometimes in the US you've had you know neo neo-Nazi forums, and across all you know countries, I think you know, um dating sites where the individual happens to be married and is it is still on them is a sort of a classic, classic thing that comes up. Uh another thing that I've noticed, uh, and this is a little bit more common, I think, um, among left-wing candidates than right-wing candidates, is you know, left-wing candidates are less likely to have you know uh more overt cases of you know um uh racism, let's say, than sometimes we see on the right, but they have more higher frequencies of violence threats that they make against people who they have disagreed with. So you can have someone, and I've seen this in my own research uh through my own clients, you know, people who are running for Congress now in the US on the Democratic side, who five, six years ago made threats of violence against people who they disagreed with online. So these are the kinds of things that you you typically retrieve.

Eric Wilson 12:56

I I wonder, uh thinking about the UK, Count Benface, uh, I'm a I'm a huge fan. I can't imagine it's easy to do Oppo research on him being from another planet.

John Artunkal 13:06

No, that's a tricky one. I I feel bad for whoever on reform has has got that on their hands. Though I would say being in a position where you have to do Oppo research on Count Ben Face in the first place is quite unfortunate.

Eric Wilson 13:19

And so, John, I've been on uh a campaign where a human researcher missed something big and and it actually it wasn't their fault, it was because the the document was removed from the place it should have been by someone who was sort of in cahoots with our opponent. Um and so you know there was someone that we could hold accountable for for that getting missed, and then they could investigate, you know, who was responsible for it. But with AI, you know, let's say you get a clean report, and then there's there's no way uh to know what it didn't flag, what Don Rumsfeld used to call the unknown unknowns. And so how do you think about the responsibility or the comprehensiveness of an AI research vet?

John Artunkal 14:10

Yeah, absolutely. Um when I first started developing AI tools for vetting purposes and opposition research purposes, I was really wrestling with this issue of um, you know, my view is a client will forgive you if you show something that isn't really a problem, it could maybe have a small chance of being a problem. But a client is very unlikely to forgive you if there was a real home run and and and you missed it, or a real red flag and you missed it. And so I ultimately think that the responsibility lies with who designs the tool, or rather, did you communicate properly to your end users what the tool does and does not do? Um so I think that it still lies with the tool, but because of that, what I've designed my platform, Argus X, to do is to be slightly more on the sensitive side. So sometimes it will pull out a post and it's debatable if there's really something there or not, but I'd rather show that than miss it, and then the client could come and say, look, that that was actually a very good tweet. Why did you miss that?

Eric Wilson 15:09

Yeah, and and that's always a tough thing in in product, right? Because you think I mean there's sort of classic examples of um aircraft design where it's it's sending too many warnings, and then you you you you miss it. And so it's um, you know, in research, you know, if you bring too many um pieces of hay and not enough needles, some stuff falls through the cracks. I I imagine that that is a very fine line you have to have to walk.

John Artunkal 15:39

It's a fine line that I spent a few hours before today doing, and after we finish speaking, I'll be going back to that uh work as well. And there's a few ways around it. So the the first one is you're always trying to improve the prompts that your AI is using to reduce the number of false positives, and we always track how many false positives we see on Argus X, and then we build a database of them. So we'll build a database of false positives so that the AI knows, okay, I flagged these kinds of posts before in the past, they're wrong.

Eric Wilson 16:08

And you you've alluded a little bit to kind of the classic things like being on a dating site when you shouldn't be threatening violence, those sorts of things. And and what's interesting is is kind of the shift that I've seen over time. So when I'm revealing my age here, I remember when social media first came uh to bear, and candidates would say, Well, you know what um why do people care about this? Or or campaign managers and consultants would say, Hey, well, what if my candidate posts something stupid? And I always point out, hey, that's not a social media problem. That's a that's a candidate problem. They're gonna say something stupid in real life if if they don't say it online. Um but you know, and we used to say a gaff was when when a candidate told the truth. Um, but now we're seeing that you know voters really gravitate to these people who are maybe committing what we would have called gaffes all the time and being very um authentic for lack of a uh a better word, um and and and posting a lot. And so I'm wondering if if we're in this I I I guess it's it's interesting because you you you point to examples where it's sort of contrary to the character, right? Like so if someone's loyal to their spouse, they shouldn't be on a dating website. That's a character flaw, versus here's a political opinion that's within the sort of normal bounds.

John Artunkal 17:26

Yeah, and I think you know the the the message to candidates shouldn't be stop posting or shouldn't it be make sure that every one of your social media posts is some milk toast platitude that doesn't really mean anything. People like authentic candidates, they like people who are a little rough around the edges. The only thing is you've got to make sure that you don't have any career enders in there, or things that you as you correctly pointed out, things that run contrary to your brand, who you say you you are. If people are authentic to themselves and there's no career enders, then social media is very powerful and they shouldn't oversanitize themselves.

Eric Wilson 18:02

And and I I think there are two ways that I encourage candidates to think about this. One is if they've ever been to like a circus and they've seen someone walk on a bed of nails. Well, you know, if you're walking around in your garden and you step on a single nail, it goes right through your foot. But if you put a bunch of them together, you you can uh stand safely on it and it makes a neat circus trick. Um similar uh thing is happening with social media. If the only times you post are these sort of um off-the-wall statements or um, you know, may maybe you're you're posting rarely, people don't get the full picture uh uh of who you are. And this is something that that comes up when we have discussions about deep fakes, right? So if you give enough people examples of what the real thing is, they're gonna know when the fake i is there.

John Artunkal 18:52

Yeah, and and uh to your point, yeah, if uh all of your posts are you talking about you know being a little bit edgy, um, then none of those are really gonna come back to bite you because that's a part of your coherence, you know, political brand.

Eric Wilson 19:06

So John, it seems to me that that text with with LLMs and AI is is basically a solved problem, right? The you know if the posts are public or or you can get access to a database, everyone's scanning them. What or or is that solved too? Where do you think if if I'm a candidate, where's my biggest risk lying right now?

John Artunkal 19:36

Well, it depends on the kind of media that you have produced over the course of your your life, right? So, you know, just because AI can read uh text doesn't mean that you're you're safe. If all we've produced is has been has been text and you've forgotten what you posted about from five, six, seven years ago. That's still where most of the big big wins come from. Um but when it comes to the analysis of uh images and videos, there's huge progress taking place there as well. So, firstly, facial recognition, big aspect of this. AI is being able to detect where your face um can be found online. The mechanism behind this, by the way, is fascinating. It essentially constructs a facial fingerprint based on the distance between your eyes and the edges of your mouth. So that even if you lost or gained 100 pounds, it would still be able to work out who you were. Um but then in terms of the content you share online and the AI analysis of that, um, that's getting easier as well. So if you've posted dozens of images on X, for example, reposts of those, the leap to analyze text to analyze images like that is becoming smaller and smaller. In fact, you can I when you use LLMs on a day-to-day basis, you can send it memes and it'll easily understand the content and if it's appropriate or not. So, you know, that's um that gap's being closed too.

Eric Wilson 20:58

Well, I want to say thank you to John Artunkel for a great conversation today about AI and opposition research. I've got uh links in the description uh where you can learn more about his work and Argus. Um check it out if that's something that that you need to have in your toolbox. And, you know, if this episode made you a little bit smarter, we just ask that you share it with a friend or colleague. You'll look smarter in the process and more people find out about the show. It's a win-win-win, and we like those around here. Please remember to subscribe to the Campaign Trend Podcast wherever you listen or watch podcasts so you never miss an episode. And you can visit our website at campaigntrend.com. We've got great newsletters over there. So if you only know the podcast, you shouldn't sign up for maybe our best practices newsletter or the rundown that goes out every Friday. With that, I'll say thanks for listening. We'll see you next time. The Campaign Trend Podcast is produced by Advocacy Content, a media producting company.