Latest / Stage Talks with Bellingcat / "Chinese influence operations and disinformation campaigns" with Jasper Hewitt
Transcript
- 0:04You're listening to a stage talk titled Chinese
- 0:07influence operations and disinformation campaigns.
- 0:10This week we were joined by DoubleThink Labs
- 0:12Jasper Hewitt as we explored the new forms of
- 0:15propaganda. We discussed spam of large operations
- 0:18in the US elections to AI influencers pushing
- 0:22state lines on TikTok, along with some tool and
- 0:25reading recommendations along the way. This talk
- 0:27was hosted by me, Charlotte Ma, on Thursday the
- 0:308th of May 2025 in the Bellencat Discord server.
- 0:50Well, for another stage talk. This week, we're
- 0:55speaking about digital influence operations and
- 0:58disinformation campaigns. An MIT study back in
- 1:012018, which feels like forever ago, found that
- 1:05fake news was 70 % more likely to be retweeted
- 1:09on Twitter than real news. And that was before
- 1:12Musk's takeover. So you can imagine what that
- 1:14looks like now. While state -led influence operations
- 1:17have happened for years, but the use of internet,
- 1:20and in particular social media, as a means of
- 1:23running information operations has grown prominently
- 1:25over the past decade. The Chinese Communist Party
- 1:28is one such political power that uses digital
- 1:31influence operations and disinformation campaigns
- 1:33to advance the party's interests in the region
- 1:35and further afield. Jasper Hewitt joins us here
- 1:39from DoubleThinkLab. DoubleThinkLab researches
- 1:42malign Chinese influence operations and disinformation
- 1:45campaigns and their impacts via the digital tools
- 1:47and methodologies they have developed. As an
- 1:50analyst, Jasper's focus has been on PRC influence
- 1:53operations during the Taiwanese and US elections,
- 1:55and he's currently investigating how official
- 1:58PRC accounts use the MetaAd library to expand
- 2:01their content's reach. We'll jump into all of
- 2:04that in a second. But first, a short reminder,
- 2:07if you'd like to ask a question, please do so
- 2:09in the chat as we talk. And please note within
- 2:13your question, if you do not want me to read
- 2:15your username out, as this is being audio recorded
- 2:18and will appear on podcast platforms. Okay. Jasper,
- 2:23before questions, can you tell us a little bit
- 2:26more about your work? And while you do so, I
- 2:29will mute. Yes, so first of all, thank you for
- 2:33that great introduction and thanks for having
- 2:35us, having Double Think Lab here. This is a great
- 2:37opportunity to talk to the whole community. So
- 2:40yeah, as she just said, Double Think Lab is a
- 2:43Taiwanese organization. We're an NGO and we've
- 2:46been founded in 2019 and yeah, we mostly conduct
- 2:49research into PRC influence operations. Now,
- 2:53I'm on the digital intelligence team, so I'm
- 2:56mainly responsible for tracking and reporting
- 2:59on all of the inauthentic accounts. And we're
- 3:02also developing automated tools to track these
- 3:05accounts because, of course, this is a very time
- 3:08consuming work. So if we can automate it, then
- 3:11we do that wherever possible. But I can talk
- 3:14about that more in the Q &A. So yeah, today I
- 3:18just want to give you first say a few things
- 3:20about FEMI or influence operations and how that's
- 3:23different from fake news and disinformation.
- 3:26And then I just want to give you a crash course
- 3:28into PRC influence operations. So we're going
- 3:31to talk about spamouflage, which is the biggest
- 3:33one. And I'm going to tell you about how spamouflage
- 3:37has been evolving over the last few years. Yeah,
- 3:42FEMI. So FEMI stands for Foreign Information
- 3:45Manipulation and Interference. It's a very long
- 3:47name, but it's basically a different term for
- 3:49foreign influence operation. Now the key term
- 3:52there is foreign. So we're mostly concerned with
- 3:55information that goes from one country to another
- 3:58and then tries to influence the political debate
- 4:02in that second country. Femi is different from
- 4:07fake news and disinformation because not all
- 4:10influence operations are disinformation. So people
- 4:14can say something that is true, right? You can
- 4:16just say like, oh, I really like that political
- 4:19candidate. And that's not necessarily fake news,
- 4:23but we care mostly about the method. That's something
- 4:27how information is spread. So we're looking for
- 4:29coordinated inauthentic behavior. So it's methods
- 4:31over over content. So if we find that there's
- 4:34like a thousand inauthentic accounts saying,
- 4:37for example, that BYD is an amazing company,
- 4:40then that would be a reason for us to to investigate.
- 4:43Now, if we look at PRC influence operations,
- 4:47there's several actors that are involved in this.
- 4:50So obviously, there's the PRC state media, right?
- 4:53This is really big and they all have they're
- 4:55all very active on social media on several platforms.
- 4:59And then there's influencers. Now, some of these
- 5:02influencers have direct ties to the PRC government
- 5:05with some influencers. It's not as clear if they
- 5:08have ties to the PRC government, but people suspect
- 5:11that they do. And then there's spamouflage. So
- 5:14that's tens of thousands of, if not more, fake
- 5:19accounts that are basically just spamming the
- 5:23internet with pro -PRC messages. So, yeah, spam
- 5:27has been defined as a massive cross -platform
- 5:30propaganda infrastructure. It was first discovered
- 5:33by Grafica in 2019. So Ben Nemo actually gave
- 5:38it the name spam because they would camouflage
- 5:42their propaganda with spam. So, yeah, you wouldn't
- 5:46only see propaganda. It would be like pictures
- 5:48of beautiful scenery or pictures of attractive
- 5:50individuals or random book quotes. And then suddenly
- 5:53there would be this PRC propaganda. Now in 2023,
- 6:00Metta actually attributed spammaflage to PRC
- 6:03law enforcement. So this was a very important
- 6:06finding, but it basically it basically proved
- 6:10what everybody was already thinking, right? Because
- 6:11all of this stuff was so pro -PRC that there
- 6:14could really be only be one entity behind it.
- 6:18Now, what is the camouflage saying? Because this
- 6:21has also changed over time. At the beginning,
- 6:23they were mostly concerned with issues that are
- 6:26sensitive to the PRC. So for example, prominent
- 6:30dissident Guo Wengui, he was targeted at the
- 6:32start, and then Hong Kong with the protests,
- 6:36COVID, and then also Xinjiang and Tibet. But
- 6:40over time, the campaign just became more overall
- 6:43pro PRC, anti -west. They started going after
- 6:46specific countries. So the USA is obviously a
- 6:49big target. Taiwan is a very big target. I think
- 6:53we're right at the forefront here in Taiwan.
- 6:56And then Japan, the Philippines, maritime disputes.
- 7:00These are all topics that Spamouflage is touching
- 7:03on. So, for example, during the US elections,
- 7:06these accounts were also very active. And they
- 7:08were basically criticizing domestic issues, but
- 7:11also US foreign policy, like, for example, US
- 7:14support for Israel, for Ukraine. And they were
- 7:17critical of both presidential candidates. They're
- 7:21also going after individuals, so specific dissidents,
- 7:25also smaller dissidents, not just the super high
- 7:28profile ones. They're all being targeted by this
- 7:30campaign. So their social media accounts, sometimes
- 7:33in the comment sections, you'll read a lot of
- 7:35hate comments. And those are also attributed
- 7:38to this camouflage. Now, what do these accounts
- 7:42look like? So I'm going to talk a little bit
- 7:44about how they operate. And then you can maybe
- 7:47also use some of that info to find these accounts
- 7:49yourself if you're interested in it. So many
- 7:52of these accounts, they would use profile photos
- 7:55of women, mostly Asian women. And many of these
- 7:58photos have been taken from small Chinese influencers.
- 8:03So I personally think this is really messed up
- 8:06because these people, technically speaking, cannot
- 8:08use Western platforms themselves. But then an
- 8:10influence operation attributed to their government
- 8:13is taking their photos to put them out there
- 8:16on these Western platforms. The accounts mostly
- 8:22swarm around the larger state media accounts
- 8:26and also influencers. So if you look at some
- 8:29state media accounts or accounts from diplomats,
- 8:32for example, Zhang Heqing is a big one. He used
- 8:35to be the diplomat to Pakistan, and now he's
- 8:37a big personality on X. And if you go through
- 8:40his posts and you look at his retweets, most
- 8:43of it will be, you will see his camouflage. Now,
- 8:48these accounts usually have a division of labor.
- 8:51So there's like cedars and then there's amplifiers.
- 8:53So seeders would post the initial content, and
- 8:56then there's like 20 amplifiers below that, that
- 9:00their only job is just to repost that. We think
- 9:04that the accounts are side work accounts, so
- 9:07to say. So they're like hardly operated by scripts,
- 9:10but they're also definitely real people behind
- 9:13it. So we know that there's real people behind
- 9:15it because they are actually most active during
- 9:19PRC working hours. So if you... plot their activity,
- 9:23you will see that's from nine to five in PRC
- 9:27working time. And then they also have a lunch
- 9:28break between 12 and two. So you can actually
- 9:31see that activity drop. So sometimes they'll
- 9:35also make robotic posts. So for example, sometimes
- 9:38at the beginning, there's robotic boot quotes
- 9:40when the accounts just start. And those also
- 9:42happen outside of working hours. Because obviously,
- 9:44when you run a script, there's no need to be
- 9:48in the office. If you think of a fake account,
- 9:52then many people might think of an account that
- 9:54was just created the other day and has very little
- 10:00followers or something. But we actually found
- 10:02that many of the, specifically the Cedar accounts,
- 10:05they were created like back in 2012, 2011, and
- 10:09they have huge followings. So we found some accounts
- 10:12that have more than 70 ,000 followers, 50 ,000
- 10:15followers. And it's not that The PRC was very
- 10:18early with making all of these accounts. We found
- 10:20that actually many of them have been hacked.
- 10:22So those were inactive accounts that haven't
- 10:26been used for a very long time. And then they
- 10:29got hacked, and then they probably got sold to
- 10:31the people running Spamouflage. So one funny
- 10:35example with an account that we were tracking
- 10:37during the US election is when we finally get
- 10:40to get all of their posts through the API, we
- 10:42found that back in 2011, this account belonged
- 10:46to a radio show in Texas. And then after that,
- 10:48it was just inactive for like 12 years. And then
- 10:52suddenly it became a spamflash account. We also
- 10:57found that these accounts appear to buy third
- 11:00party, like retweet amplification, especially
- 11:04on X. So during our investigation, we found over
- 11:08a thousand accounts that artificially boost these
- 11:11spamflash posts. And these accounts were actually
- 11:15crypto accounts and Web3 accounts that normally
- 11:18speaking have a lot of, normally speaking, all
- 11:22of their reposts are of like large crypto accounts
- 11:25like ChainGBT or London Real TV. So it was really
- 11:30interesting to see that. in those timelines,
- 11:33suddenly you would see these spam of flash posts
- 11:35popping up. So there's some interesting dynamics
- 11:37going on there. It's not like they do everything
- 11:39in house. There's definitely some collaboration
- 11:42going on with spam companies, potentially also
- 11:46from other countries. Now, spam of flash is notoriously
- 11:50ineffective. So many researchers say that spam
- 11:53of flash has failed to get a lot of authentic.
- 11:59authentic interaction over time. So yeah, what
- 12:02I just said, they buy their retweets, many of
- 12:05their followers and their retweets or other spam
- 12:07of flash accounts. So yeah, should we really
- 12:09be caring about this? Well, today I'm here to
- 12:12say that, yes, I think we should keep caring
- 12:14about spam of flash because it is evolving and
- 12:18they are definitely trying to find new ways to
- 12:22become more relevant. So I'm now going to share.
- 12:25two or three examples of what we found in which
- 12:29we can see that spam of flash is becoming more
- 12:31relevant, I think. So during the U .S. elections,
- 12:35they were continuing what they were always doing,
- 12:38like posting a lot of pro -PRC stuff and saying
- 12:40some critical things about the United States.
- 12:43But we also saw something new. And this was called
- 12:46MegaFlash. So there were these spam of flash
- 12:49accounts that were impersonating Trump supporters.
- 12:52And this was first discovered by Elise Tomas
- 12:55at the Institute for Strategic Dialogue. And
- 12:59yeah, there were these accounts that really pretended
- 13:02to be American. So they had like in their bio,
- 13:05they had like all these American flags and said
- 13:07like stuff like hashtag NRA and stuff like that.
- 13:13They were interesting because it was harder to
- 13:16prove that they were connected to spamaflash.
- 13:18So whereas normally speaking, spamaflash is very
- 13:21straightforward. They just post a lot of pro
- 13:23-PRC stuff. So it's very easy to kind of see
- 13:27where their allegiance lies. But then with these
- 13:31accounts, they didn't post any pro -PRC stuff.
- 13:33They just posted pro -Trump stuff, which was
- 13:36really weird. We were tracking a bunch of these
- 13:39accounts and we already had the idea that maybe
- 13:42these accounts were connected to the campaign
- 13:44because some of the crypto accounts were also
- 13:49amplifying these accounts. And then their posting
- 13:53times were also in PRC working times, even though
- 13:56they all claimed to be Americans, which is weird
- 13:58because that would be the middle of the night
- 14:00in the United States. And then, at some point,
- 14:04they made this... So, Sprimal Flies is very clumsy.
- 14:06They sometimes make very clumsy mistakes. And
- 14:08then, at some point, like seven out of the 15
- 14:11accounts that we were tracking, they suddenly
- 14:13posted this account in Chinese criticizing the
- 14:17Falun Gong, which was just completely out of
- 14:19context of what they were doing before. And I
- 14:21think after that, they kind of realized that
- 14:24the batch was compromised. They just left those
- 14:26accounts and they haven't been used since. So,
- 14:29these accounts are still online. If you go through
- 14:31our US election report, then you can find the
- 14:34account name. Now, another thing, one specifically
- 14:38interesting account, it goes by the name of Edward
- 14:41D. Mayer. It's also still going today, even though
- 14:45we've reported on it many times. This account
- 14:49actually became really relevant. I think it's
- 14:51been the most successful spam camouflage account
- 14:53that we've been tracking over the last few months.
- 14:56So this account actually gamed the whole X algorithm
- 15:00by posting Street Fight videos. So instead of
- 15:03boring spam, like some scenery, this account
- 15:07found that if he posts Street Fight videos, then
- 15:10the algorithm just boosts his posts. So he got
- 15:13some Street Fight videos for which he got like
- 15:169 million views. And this kind of spilled over
- 15:20to some of the propaganda posts that he would
- 15:22make. So on some of the propaganda posts, he
- 15:25would get like 100 ,000 views. Now, this is a
- 15:28very weak spillover, but still a very big difference
- 15:31from what Spamouflage usually gets in terms of
- 15:34interaction. So with the other Spamouflage accounts,
- 15:37we maybe look at like 10 ,000 views, 15 ,000
- 15:40views on a good day, and then 20 ,000 views would
- 15:42be like the jackpot, at least for what we check.
- 15:46But this account really managed to game the algorithm.
- 15:49and found a very interesting way to become more
- 15:54relevant. At some point, this account got a warning
- 15:58label to its account from X saying that it was
- 16:02posting violent content and that influenced how
- 16:04it got recommended by the algorithm. After that,
- 16:09the account became irrelevant again. But he still
- 16:13continued like he created a few new accounts
- 16:15that I think were done by the same operator.
- 16:19And he also created like an X community, which
- 16:22was also interesting to see sort of trying out
- 16:24all these new functions. And this account was
- 16:27also actively replying to authentic users with
- 16:31large following. So he was just like talking
- 16:33to some people and saying stuff about immigration
- 16:36and then people would actually respond to it
- 16:38and go into into dialogue with it. So, yeah,
- 16:43this was a very interesting finding, and it shows
- 16:45that Spamouflage, they don't sit still and stay
- 16:48irrelevant. They really look for ways to become
- 16:51more relevant. Now, another thing that they've
- 16:53been doing is experimenting with artificial intelligence,
- 16:57of course. So I'm happy to talk more about that
- 17:00in the Q &A, because it's a big topic. But yeah,
- 17:05another thing that they've been doing is experimenting
- 17:08with new platforms. So there's a lot of new platforms
- 17:11popping up, right? Blue Sky, Threads, and especially
- 17:15Threads is huge in Taiwan. I think for now, for
- 17:18young people, Threads is probably the most important
- 17:20platform in Taiwan. And Taiwan is going through
- 17:25some political challenges right now. Yeah, there
- 17:31have been a lot of influence operations from
- 17:35spam of flash targeted at Taiwan, specifically
- 17:37at the ruling party right now. So in this case,
- 17:42we also see a similar pattern. So also the spam
- 17:45of flash pattern, like spam and then propaganda.
- 17:48But it's more like the mega flash. It's not overly
- 17:51pro -PRC because they know that most Taiwanese
- 17:54people don't necessarily associate with that.
- 17:56So it's more toned down. It's more like with
- 17:59the mega flash. They're only posting specific
- 18:02political messages that they think people might
- 18:06associate themselves with. And some of these
- 18:10posts actually already reached more than 80 ,000
- 18:13views and 9 ,000 likes on the political posts.
- 18:16So this is very, very high compared to what we
- 18:19saw on X. This might also be related because
- 18:22the threats algorithm just works differently.
- 18:26And yeah, on Blue Sky, we've also seen a few
- 18:30spam of large accounts popping up. And Blue Sky
- 18:32is particularly interesting because it has a
- 18:34really good open API. I don't know if anyone
- 18:36played around with it yet, but Blue Sky is just
- 18:40the ultimate platform to create an army of bots
- 18:43and just let it run. So yeah, but nobody is really
- 18:47using it yet or not enough people are using it
- 18:50yet. So if everybody switches to Blue Sky tomorrow,
- 18:54that would be very interesting. And also, It's
- 18:56also a lot easier to do analysis on Blue Sky
- 19:00because the biggest challenge, I think, in this
- 19:03work has been that all of the platforms are closing
- 19:05down their access to researchers. So CrowdTangle
- 19:10disappeared, right? Got replaced by the content
- 19:12library, but it's not the same. And the X API
- 19:16is, yeah, basically completely closed off, right?
- 19:20You have to pay a lot of money. I think if you
- 19:22want to get a million tweets per month now, you
- 19:25need to pay five thousand USD a month. And in
- 19:27the past, a researcher could get two million
- 19:30tweets per month for free. So, yeah. And TikTok
- 19:37is also a challenge in itself because the video
- 19:39format is just, yeah, kind of a pain in the ass
- 19:44to find information, especially because their
- 19:46search function is very incomplete. If you just
- 19:48typed like. if you do a keyword search, then
- 19:50it just gives you like 10 random videos or 20
- 19:53random videos from random years that it thinks
- 19:55you might be most interested in. So it's very
- 19:58difficult to do like exhaustive research there.
- 20:03So yeah, I think that's it for me and we can
- 20:06go to the Q &A. Amazing. Thank you. Yeah, I think
- 20:11most people within this audience or at least
- 20:13in the Bellicat staff side have experienced that
- 20:17frustration with the the tools closing down to
- 20:21monitor social media sites. Before going to audience
- 20:24questions, I was intrigued actually to ask you
- 20:27what the differences are, particularly as we've
- 20:29got such a research -based audience. What's unique
- 20:33about researching Chinese disinformation campaigns
- 20:36in comparison, for example, with US state -led
- 20:40disinfo or Russian state -led disinfo campaigns?
- 20:44Are there any unique obstacles or anything that
- 20:48researchers who are used to looking at, for example,
- 20:52Russian propaganda might be surprised to encounter
- 20:54when it comes to Chinese propaganda. Yeah, I
- 20:59have to be really honest in that I don't know
- 21:03enough necessarily about the Russian disinfo
- 21:06campaigns. This is something that I know that
- 21:08I should get more into, but I keep postponing
- 21:11it. I think perhaps the language element, the
- 21:17language aspect is very unique, maybe, to the
- 21:22Chinese influence operations. In the beginning,
- 21:26many of these accounts would post a lot in Chinese,
- 21:30but then now they're slowly using more English.
- 21:33Also, now there's AI, so it's more easy to just
- 21:37translate, oh, so there's less language mistakes.
- 21:40But if you're looking at influence operations
- 21:43from the PRC towards Taiwan, which is interesting
- 21:46because they both technically use the same language,
- 21:48right? But then there's a difference in the characters
- 21:51that Taiwan and China use. So Taiwan uses the
- 21:54traditional characters and China uses simplified
- 21:56characters. look for in posts that we think might
- 22:02come from the PRC is we're looking for mistakes
- 22:06in the conversion in the conversion between those
- 22:08characters. So many PRC threat actors, they will
- 22:11write a post in simplified and then they'll throw
- 22:14it through software to convert it to traditional
- 22:17characters. And then there's always mistakes
- 22:19there. So, yeah, we're kind of like always looking
- 22:23for these simplified characters that might give
- 22:25us a hint. Yeah, absolutely. You can often find,
- 22:32for example, with Russian disinfo campaigns as
- 22:36well, language, even in translated English, can
- 22:41help you spot if something is potentially misleading
- 22:44or not, because often the language translation
- 22:46isn't incredible and, as you mentioned, is often
- 22:49based on translation bots online. We've got quite
- 22:54a few questions coming through. We've had a few,
- 22:58because you mentioned about the nature of social
- 23:01media as well. Sia Skippy asked, has X's moderation
- 23:06of disinformation varied with the nation of origin
- 23:09and content of the disinformation? As in, do
- 23:11different countries get different and different
- 23:14governments get different treatments? And kind
- 23:17of mentioned with that, Ozkom said, you've mentioned
- 23:19that you've tried to get these accounts banned
- 23:21with little effect. What other mitigation opportunities
- 23:23do you see? So when you're looking at social
- 23:26media, Does moderation change depending on where
- 23:30these disinformation campaigns come from? And
- 23:34how what other mitigation opportunities do you
- 23:39see when banning these individual accounts doesn't
- 23:42work? Okay, so I'm not sure if there's if there's
- 23:45a difference in terms of in terms of geography,
- 23:49like if you do it in this country, or if you
- 23:50do it in other country, what I did notice is
- 23:53there's a difference in time. So if we find an
- 23:57account now and we write on it, I really highly
- 23:59doubt whether X will ever delete it. But when
- 24:03we were doing this research during the US elections,
- 24:05like in the weeks leading up to the actual election,
- 24:08and we had a collaboration with Voice of America.
- 24:10So Voice of America was just churning out articles
- 24:14about this all the time. And back then, we did
- 24:17notice that a lot of accounts that they reported
- 24:20on actually got deleted, like in the days after
- 24:23the article broke. And in some cases where we
- 24:28really found that someone was being impersonated,
- 24:30for example, and then an ex -spokesperson actually
- 24:34came out to say something about this and said
- 24:37that they deleted that specific account. So I
- 24:41guess at that point, even ex itself realized
- 24:45that they had to do something. because otherwise
- 24:47there might be too much pressure on them that
- 24:49they're not doing anything. And there's other
- 24:54accounts like, for example, the pay for repost
- 24:58accounts that I mentioned, like the crypto accounts,
- 25:00they were banned very, very rapidly. But I think
- 25:03that's just like an anti -spam campaign that
- 25:05they always have running and it's not necessarily
- 25:07related to disinformation. But like those thousands
- 25:10of accounts that we found back then, I think
- 25:12all of them have been deleted today. So in terms
- 25:17of other methods for mitigation, if the platform
- 25:22doesn't want to take it down, I don't really
- 25:31see another thing that you can do. Maybe if a
- 25:34lot of people report it, maybe at some point
- 25:37they will do it. I also am very skeptical about
- 25:40that. So I think the best thing you can do is
- 25:42to just tell other people about it so that everybody
- 25:45knows that this kind of this badge exists and
- 25:48that people are aware that these kind of accounts
- 25:51belong to an influence operation. And that's
- 25:54especially relevant right now for Taiwan, where
- 25:57so many people are on threats. And right now
- 25:59there's so many disinformation going on on threats
- 26:01and normal people are seeing it in their daily
- 26:04lives and are talking about it at work. Yeah,
- 26:08I think just calling it out and calling out that
- 26:10this exists can sometimes already be enough.
- 26:14Absolutely. We have seen a pattern personally,
- 26:18though, at Billingcat, for example, and I know
- 26:20a couple of other news organizations have as
- 26:22well. When it comes to news impersonation, when
- 26:26there's tactics where they're impersonating news
- 26:29organizations, for example, calling it out has
- 26:32actually given these individual accounts more
- 26:37viewership and more engagement. So then they've
- 26:40been using that as a tactic. And we've seen an
- 26:42outtake in, for example, in terms of the Russian
- 26:47disinformation context, people, Russian disinfo
- 26:51drivers, emailing and trying to get the attention
- 26:55of European fact -checkers, for example, so that
- 26:58they can fact -check the content to then drive
- 27:00more engagement their way. Interesting model
- 27:04that has been working for them and is something
- 27:06that we've spoken about in a stage talk before.
- 27:11We've got so many questions rolling in. I'm hoping
- 27:14I can keep up with them all. Just keeping on
- 27:17the social media side and as you were talking
- 27:20about the spammer flage, particularly within
- 27:22the election that you spoke about, Skirskippy
- 27:25asked earlier as well, how many of those followers
- 27:29of these accounts are themselves legitimate or
- 27:34bots? For example, my experience on Twitter is
- 27:37that accounts just naturally accrue totally authentic
- 27:39followers at a steady rate. Is there kind of
- 27:41a secondary group of bots that are being created
- 27:45to amplify those particular tweets or those particular
- 27:49posts? Is that part of the plan? Or are these
- 27:54real people kind of getting suckered in to kind
- 27:58of the propaganda and the messaging? Yes. So
- 28:01this is a very important question. And I don't
- 28:04know if you're familiar with the breakout skill.
- 28:06So this was also developed by Ben Nemo. So the
- 28:10breakout skill, basically, when you look at disinformation
- 28:13or influence operations, you want to see like,
- 28:16did this post break out of its initial bubble.
- 28:19And the initial bubble is just an inauthentic
- 28:22bubble with just a bunch of fake accounts that
- 28:25are following each other and are liking each
- 28:28other. And then if it breaks out of that bubble,
- 28:31you would be in category two. And if it breaks
- 28:34out on multiple platforms, you'd be in category
- 28:36three. And then it just keeps going up. So we're
- 28:39always trying to look for if these accounts are
- 28:42getting authentic interactions from real accounts.
- 28:48And it's difficult to look at all the followers.
- 28:52It's possible, but if you have these large accounts
- 28:55that I mentioned earlier, like these hijacked
- 28:57accounts, those are tens of thousands of followers.
- 29:00I think it might be somewhat of a waste of time
- 29:02to go through all of their followers. So what
- 29:06we usually do is we look at the comment section.
- 29:09Because the comment section of spammer flash
- 29:11accounts are very, let's say, Most of the other
- 29:16spam and flash accounts that comment on these
- 29:17posts, they all agree. They all say stuff like,
- 29:20yes, yes, good point, stuff like that. It's very
- 29:22kind of like monotone. But if at some point you
- 29:25see replies that disagree with the post, or you
- 29:30maybe see replies with people saying stuff like,
- 29:33oh, this is weird, or whatever for some reason,
- 29:36then you can assume that this post has somehow
- 29:39reached authentic users. So I think that's the
- 29:42easiest way to see that. Yeah. And then, yeah,
- 29:49just a little more on that. So I think they themselves,
- 29:53Spammouflage itself is creating more and more
- 29:55Spammouflage accounts to then follow all of these
- 29:57accounts and also repost them. And then on the
- 30:01other hand, I think they're also buying it from
- 30:03third party. So like spam companies, they're
- 30:06also buying likes and reposts. You mentioned
- 30:13that in terms of the propaganda aimed at Taiwanese
- 30:18population, that's happening a lot on threads.
- 30:23We've actually had a question from Sohan D'Souza
- 30:25who's asked, would you please advise regarding
- 30:27optimal techniques for threads investigations?
- 30:30The advanced search doesn't seem too bad, but
- 30:32transparency even via external services seems
- 30:35lacking. Could you speak a little bit about how
- 30:38the difficulties and the ease of investigating,
- 30:43particularly on threads, is that's a new platform
- 30:45that many of the researchers, for example, in
- 30:47the audience may be new to and want advice on.
- 30:52Yeah, sure. So I think one thing that's difficult
- 30:56about threads is that everything is new. So sometimes
- 31:00when people look for fake accounts on X because
- 31:04in most cases, it still are like new accounts,
- 31:06right? So you would suddenly see an account that
- 31:08was created like three months ago saying all
- 31:11this stuff. But then like on threads, it's not
- 31:13weird to have an account that was created three
- 31:15months ago because everybody is just joining
- 31:17this year. But yeah, I think threads is available
- 31:21in the content library now if you have access
- 31:23to that, but it has a lot of limitations, like
- 31:25only certain accounts you will see. So it's not
- 31:28very helpful. What I think is a very useful tool
- 31:33is there's this add -on that you can get in the
- 31:37Chrome store. It's called Single File. And it
- 31:40basically just lets you get a .html file of the
- 31:45whole page that is currently loaded. So you can
- 31:48just go to someone's timeline, and you can scroll
- 31:50all the way down, and you can click the add -on.
- 31:54And then it will just download a .html file of
- 31:57that whole. of that whole page. So that would
- 32:00include all the posts, all the images. And what
- 32:04you can also do after that is you can write a
- 32:07script to organize that data into a data frame.
- 32:10So technically speaking, you're not scraping.
- 32:12You're not using Selenium and stuff. So Facebook
- 32:15won't detect you and ban your account, because
- 32:19you're just using this tool to get all of the
- 32:21HTML. And then you can do with that data what
- 32:24you want after. Another thing that's difficult
- 32:29on threads is I found that sometimes certain
- 32:34things aren't really working. Like when I try
- 32:37to click the account creation date, for example,
- 32:41on my laptop, sometimes I get a lot of errors,
- 32:44but then when I do it on my phone, it works.
- 32:47So I feel like the app works better on your phone
- 32:49than it works on your computer, but I'm not doing
- 32:51investigations on my phone. And then you also
- 32:55have to do a lot of clicks to get there. So it
- 32:57just takes a lot of time. Other than that, I
- 33:01think I like that it has a keyword search function
- 33:05with a recent tab. So I think it looks pretty
- 33:09exhaustive like what you have on Twitter. So
- 33:12I'm really happy about that. That it's not like
- 33:16TikTok. Thanks for that little tip. I've linked
- 33:21that in the chat, and I've personally just gone
- 33:24and downloaded it. I have a similar one called
- 33:26go4page, but that just takes a PNG, so it's good
- 33:30to know that there's one for HTML. To say more
- 33:33on that, this works for almost every website.
- 33:36The only downside is that the .html files are
- 33:39very big, so if you get someone's full timeline,
- 33:41you might get 100 MB. But yeah, I think it's
- 33:47kind of like a half manual, half automated way
- 33:49for data collection. It's also a great way to
- 33:51archive, right? We all know the trouble of trying
- 33:54to archive something with your VPN on and then
- 33:56you have to, it just takes ages. But with this
- 33:58thing, you can just click it and save it for
- 34:00later. And then if the account gets deleted,
- 34:01you can just go back to the .html and make your
- 34:04screenshots from there. And that's a great time
- 34:08for me just to shout that we have a Bellingcat
- 34:11auto archiver, which has recently had some updates
- 34:15done to it. So please do check that out as well.
- 34:18Just linked the guides that in the chat. We've
- 34:23had quite a few questions and I'll let you take
- 34:26a... Choc? Cool name. It's the most recent question
- 34:30to kind of encompass this, but we've had quite
- 34:33a few questions about how you spot Chinese Finney
- 34:36or Spammelfarge accounts when their activity
- 34:38seems to be non -discernible from other Finney
- 34:42without an in -depth analysis. Are there any
- 34:45telltale signs to spot Spammelfarge accounts,
- 34:51specifically Chinese Spammelfarge accounts? Yes,
- 34:55so I think some of the things that I just mentioned,
- 34:59like for example, that they're mostly active
- 35:02during PRC working hours, that is one that you
- 35:05can always check. And you can also look at the
- 35:13content. So for example, I mentioned a few things
- 35:15like specific dissidents or Hong Kong or Xinjiang.
- 35:20So these are like... target topics that they
- 35:23always focus on. So if it's like a traditional
- 35:26spam -a -flash account, it will talk about these
- 35:28topics and it's very easy to spot. So you can
- 35:31just go to, for example, you can go to People's
- 35:34Daily's X account and then just go through its
- 35:38repost or look in the comment section and you'll
- 35:40find a bunch of spam -a -flash accounts. Now
- 35:43it becomes more tricky if the spam -a -flash
- 35:45accounts are like these mega -flash. type of
- 35:48accounts, right? So they're not posting all of
- 35:50this traditional pro PRC stuff. And you might
- 35:52think like, oh, this could be, this could be
- 35:54anyone running this influence operation. So how
- 35:58we attributed those accounts to spamouflage is
- 36:02by looking at the people who amplify it. So that's
- 36:05where these kind of weird crypto accounts came
- 36:07in. So we noticed that these accounts were amplifying
- 36:11a lot of spamouflage and then suddenly like right
- 36:14before the election, like five days before the
- 36:16election, just all these new accounts were created
- 36:18that criticize specific lawmakers. And they only
- 36:22had three posts and nothing else. But when we
- 36:25looked at the reposts, we saw that there was
- 36:28actually quite some overlap with the accounts
- 36:30that amplify the traditional spam of flash and
- 36:33also PRC state media accounts. So that's how
- 36:36we were able to attribute it. There's also more
- 36:39on that in our US election observation report,
- 36:43if you're interested. And I highly recommend
- 36:46anyone to just read a lot of papers about spamouflage,
- 36:50and you will see a lot of account characteristics
- 36:53that you can use to find them. And once you find
- 36:57a few, you can just keep snowballing. Someone
- 37:00asked earlier actually in the chat, in the election
- 37:05coverage that you saw, were they... only promoting
- 37:08Trump rhetoric, or was it a little bit more broader
- 37:16across the political spectrum? They also asked,
- 37:18have you seen any similar campaigns targeting
- 37:22European politicians, for example? For example,
- 37:25last year was a huge election year across the
- 37:27world. Have you seen any similar trends in terms
- 37:31of European elections? So first, on the first
- 37:36question, most of the traditional spam camouflage
- 37:41accounts were anti -everything. They did say
- 37:45a few things that maybe aligned more with Trump.
- 37:48So they were, for example, a bit more, well,
- 37:51they're anti -U .S. support for Ukraine, which
- 37:55maybe you think that, okay, maybe that's more
- 37:57closely connected to what Trump is saying. There
- 38:00were also critical of the LGTBQ movement. So
- 38:03maybe you can also draw some connections there.
- 38:07But with the mega flush account, specifically
- 38:09the accounts that are really pushing Trump, we
- 38:12only found that for Trump. We didn't find similar
- 38:15accounts that were supporting Biden. One interesting
- 38:20thing, though, is that basically all of the spam
- 38:23flush accounts that we were tracking were very
- 38:25anti LGTBQ. And then like those five days before
- 38:30the election, when these accounts were created
- 38:32that criticize specific lawmakers, they were
- 38:35criticizing Republican lawmakers for being anti
- 38:38-LGBTQ. So that's just a completely different
- 38:41standpoint kind of playing both sides. So that
- 38:45was it was interesting to see that. And for Europe,
- 38:49I haven't really been tracking the European election
- 38:54season and also the accounts that I was tracking,
- 38:57it was more quiet about the European elections,
- 39:01but I'm pretty sure that some stuff did go on.
- 39:05Maybe that's something our community can have
- 39:07a look into. Something to track, guys. A little
- 39:11homework for you all. You mentioned earlier that
- 39:15often they use images of influencers. as part
- 39:22of the accounts, they might not be aware, these
- 39:24influencers might not be aware that their images
- 39:27are being used or it's being used without their
- 39:30consent or their control. Do you also see a trend
- 39:33of proxy actors or self -interested actors helping
- 39:36to boost propaganda narratives in Asia, particularly?
- 39:40For example, social media influencers peddling
- 39:42Chinese state lands. I think on your website,
- 39:46it might attend economic capital owners or suitors.
- 39:50Is there any pattern to there to what these influences
- 39:53look like, say, or do they kind of transcend
- 39:57algorithmic communities if they do exist? Yeah,
- 40:01so one of our partners actually wrote a big article
- 40:05on that. That's also on our medium. And she found
- 40:10that on WeChat, so now you have these WeChat
- 40:13stories, and she found that there were a lot
- 40:17of influencers. Chinese and also Taiwanese influencers,
- 40:22like living in Taiwan, who would then post a
- 40:25lot of very pro PRC narratives. So to say like,
- 40:29oh, Taiwan should join China and that kind of
- 40:33stuff. But the conclusion there was that she
- 40:38believes that they were only saying these things
- 40:40because they know that they might be boosted
- 40:43by the algorithm for saying that. And then they
- 40:45were all running these online shops on the side.
- 40:49And this is interesting because we found a similar
- 40:52thing on TikTok. I think Voice of America also
- 40:56wrote an article about that, where we found these
- 41:00accounts that were kind of operated, seen from
- 41:03out of Singapore, and they were AI avatars, often
- 41:07of Western -looking women or Russian or Ukrainian
- 41:11women. I think that's how it kind of started.
- 41:14And they would start by giving this pro PRC propaganda
- 41:18talk and then like halfway through the video
- 41:20they would suddenly start promoting these vitamin
- 41:22products and it's just so weird but there's this
- 41:27this theory that these videos were maybe originally
- 41:29made for PRC social media and they think like
- 41:33oh if we post this kind of stuff the narrative
- 41:34that the algorithm will push us because that's
- 41:37what That's what Xi Jinping wants us to say.
- 41:40And then we use that to then sell our products.
- 41:44But this is still kind of like a hypothesis.
- 41:46But we did find this in multiple different places.
- 41:49That's really interesting. Yeah, I haven't seen
- 41:54it linked with political comment, but I've definitely
- 41:58seen news influences. fake ones, using a news
- 42:03hub to then go and sell a well -being product,
- 42:06for example. It's really interesting that that's
- 42:09a tactic that they might be using, particularly
- 42:11on TikTok. Thanks for sharing. And I've linked,
- 42:14I think, is it called Patriotism for Sale, the
- 42:16WeChat one? Yes, yeah, that's the one. I've linked
- 42:20that within the chat. Do check out DoubleThink
- 42:24Labs' medium, by the way. It's full of incredible
- 42:28information. Hania Reyes just put, I've seen
- 42:32hacked influencer accounts start to sell products
- 42:34on TikTok too. Mashtar said maybe they modeled
- 42:38some of it off Alex Jones info walls as they
- 42:42were pushing vitamins. Somebody's just dropped
- 42:46your DoubleThinkLab medium in the chat as well.
- 42:48That's really cool. Thank you for doing that.
- 42:51Still so many questions coming in. You mentioned
- 42:54the big... the big term AI in your last answer,
- 43:00could you tell us a little bit more about how
- 43:02AI is being used in influence operations going
- 43:07forwards and is that a new phenomenon or has
- 43:10it been going on for a few years? I think right
- 43:15now the thing that AI, that people are most afraid
- 43:20of with AI are not are not really happening yet,
- 43:24at least with PRC influence operations. It feels
- 43:26like they are kind of experimenting with it.
- 43:30So I kind of divide it up for myself. I think
- 43:33AI is being used in four different ways. Now,
- 43:35the first one is AI that is being used to deceive,
- 43:38right? So you're trying to create photos or videos
- 43:41that look real, but actually they're fake. And
- 43:43you're really trying to fool people. I changed
- 43:49their perception of things through these fake
- 43:51videos or photos. And so far from the PRC influence
- 43:55operations, I have not really seen that. Most
- 43:59of the use of AI has been AI to convey a certain
- 44:04idea. So they would, for example, create an AI
- 44:09generated image of, I don't know, like Trump
- 44:13and Netanyahu kissing or something like that.
- 44:15And that would be like everybody knows that that
- 44:18this is fake, right? It's very obvious, but it's
- 44:20being used to get this idea out there, get this
- 44:26visualization. I guess it's just a replacement
- 44:27for a cartoon if you don't have time to draw
- 44:30a cartoon yourself. And then there is another
- 44:34method, which is also interesting, and that's
- 44:36AI as a technical assistant. So for example,
- 44:41we found Also, in the US election, we found that
- 44:45some accounts, like a few days before the election,
- 44:48they suddenly uploaded a video and it was like
- 44:52speech in English and then some stock images
- 44:56in the back. But we actually found that the speech
- 44:59originates from a Baidu post in Chinese, like
- 45:03an essay that someone posted. And yeah, we highly
- 45:07suspect that they used AI to translate and then
- 45:09also use text to speech AI to create the voice
- 45:13to say this. So this opens up a lot of possibilities.
- 45:17Right. So they can now just get any any data
- 45:20from the Chinese cyberspace and just turn it
- 45:23into a spoken like perfectly spoken English audio
- 45:27fragment and just put it on Western social media.
- 45:31There was something similar in Taiwan during
- 45:34the election, so someone made a PRC threat actor
- 45:38wrote this book about Tsai Ing -wen, the previous
- 45:41president, and they called it The Secret History
- 45:44of Tsai Ing -wen, which is just a book full of
- 45:47nonsense about Tsai Ing -wen. And then they got
- 45:49all of these AI avatars that they got made with
- 45:53CapCut, so I think that belongs to ByteDance,
- 45:55and they made all these AI avatars. to narrate
- 46:00the book in kind of like a newsroom setting.
- 46:04So yeah, just kind of using these AI tools to
- 46:08very quickly get certain information out there.
- 46:12And I think the most scary one, which is like
- 46:16AI for full automation, which is the thing that
- 46:19I am most afraid of, we've seen accounts experimenting
- 46:22with it. Like we've seen posts, even posts by,
- 46:25for example, Zhang Heqing, like the former the
- 46:28former PRC diplomat. He has a lot of posts that
- 46:31look a lot like they were AI generated. And I
- 46:34think that that's where it becomes really scary
- 46:36because that's the moment where you can just
- 46:38create 10 ,000 bots and give them a wide range
- 46:43of prompts. You can give them guidelines and
- 46:45they just can create original content all the
- 46:48time and it just won't stop. And especially because
- 46:52OpenAI, like in February, they released their
- 46:55newest report in which they found that PRC threat
- 46:58actors were using JetGBT to create critical comments
- 47:03of certain dissidents. Then you can just unleash
- 47:09an army of 10 ,000 bots on one person, especially
- 47:13if you have a model that you can locally run.
- 47:16This is what I'm most afraid of, but it hasn't
- 47:19really happened yet. Thanks so much for explaining
- 47:24some of the tactics that they take in terms of
- 47:27AI. In regards to other questions, Mini Joyce
- 47:32has asked, what are the odds China uses cyber
- 47:35capabilities to bolster their Femi operations?
- 47:38I mean, more than they already do. Is there a
- 47:40sense where they're using cyber capabilities
- 47:45to bolster the reach of these kinds of operations?
- 47:49I think that is definitely possible at some point,
- 47:53especially if If you start looking at full automation
- 47:58that I was just talking about, I think at that
- 48:00point, they could definitely tap into some of
- 48:03the cyber resources that they have. But I also
- 48:08think that you would maybe see this on creating
- 48:14news websites and stuff like that. I think you
- 48:16would see it would be more prominent there than
- 48:19on social media accounts. And yeah, of course,
- 48:22they have Yeah, you have the rise of PRC platforms
- 48:27right now. So yeah, TikTok and also Xiaolongsu,
- 48:32like Red Note. People have also started downloading
- 48:35that. And of course, there's also a lot of, I
- 48:38think, embedded influence operation going on
- 48:41within these apps with pushing certain content
- 48:44and suppressing other content in the same way
- 48:48that happens on Weibo or like the really proper
- 48:52PRC platforms. So they're definitely using their
- 48:56cyber capabilities there. As you asked earlier,
- 49:00and this goes back to kind of people in the audience,
- 49:03for example, who might be interested in getting
- 49:05involved in all of this. What research, what
- 49:08resources slash learning platform slash certifications
- 49:12would you recommend someone trying to get into
- 49:15the field of FEME? I think the field of FEME
- 49:19is kind of Well, it's maybe similar to OSINT
- 49:25in general, but the field of FMI, most of the
- 49:29analysts, especially at a double think lab, all
- 49:32come from kind of like a political science background.
- 49:34I also have like a background in Chinese studies
- 49:38and Asia Pacific studies. And then I learned
- 49:41some coding and some data science, but we also
- 49:44have an engineering team that can really help
- 49:46us with more of the technical things. But I think
- 49:51just some basic Python, especially pandas, could
- 49:55be very useful. Because, for example, when we
- 49:58write algorithms to detect coordinated inauthentic
- 50:03behavior, we look a lot at, for example, post
- 50:05similarity. And the time between posts is also
- 50:10very important. So if you can very quickly. kind
- 50:14of like manipulate the data to find similar posts
- 50:17using certain Python packages and also find posts,
- 50:20batch posts that were done in very rapid succession,
- 50:24then that is definitely very helpful. I guess
- 50:26if you're very good with Excel, you can also
- 50:29do all of that stuff. I don't necessarily think
- 50:33you need any certification. or specific certification.
- 50:40At least I don't really have that. But I do recommend
- 50:45that you just read as much research as possible
- 50:49from, for example, DoubleThink Lab or DFR Lab
- 50:52or Microsoft Threat Intelligence. You also have
- 50:56Google is also doing research in all of this.
- 50:59There's so many parties out there that have researched
- 51:01spamouflage before. And yeah, you can kind of
- 51:06imitate or copy their research methods. And as
- 51:11a reminder, you can always ask for support and
- 51:14help within this survey as well. lots of channels
- 51:18for that, which I'm sure the mods can point you
- 51:20to. Particularly when it comes to covering China
- 51:24specifically, we have region -specific channels
- 51:27as well where you can debate and talk about the
- 51:30current news and current investigations. We've
- 51:34also had a recommendation by somebody to check
- 51:37out Sam Cooper's work. We'll go on Chinese interference.
- 51:39There's a Canadian journalist who's uncovered
- 51:41quite a bit about Chinese influence and interference
- 51:46in Canada specifically. Do you, Jasper, do you
- 51:50have any other recommendations for investigative
- 51:52journalists perhaps that our team, our audience
- 51:55could look at? I'm sorry, what do you mean precisely?
- 51:59In regards to journalists within the space as
- 52:01well, as you mentioned the organizations that
- 52:04are working within the space or any journalists
- 52:06that also stand out to you that are working within
- 52:09this space and highlighting Chinese influence
- 52:11operations. Yeah, so I've worked very closely
- 52:15together with Wenhao Ma from VOA. So, yeah, he
- 52:22has a sub stack and I'm sure he will lend a nice
- 52:28job again soon and he will keep reporting on
- 52:30it. But I think his China reporting has been
- 52:33really good and he very often also touches on
- 52:36issues related to influence operations. So I
- 52:40definitely recommend following him. And then
- 52:43there is some other, it's not really journalists,
- 52:47but it's more like researchers also. So it's
- 52:50called discourse power. It's also this sub stack.
- 52:54And they also write a lot of excellent stuff
- 52:57that also touches a lot on PRC influence operations,
- 53:01but also PRC propaganda in general. Amazing,
- 53:06getting so many reading recommendations right
- 53:08now. Thank you. I've put both of those within
- 53:11the chat. We only have five minutes left, so
- 53:15I just want to make sure that we've covered everyone's
- 53:19questions. We had a question earlier particularly
- 53:22about the influence of Chinese operations or
- 53:27digital operations in Africa. Are you familiar
- 53:31with any large scale influence campaigns by the
- 53:37PRC, for example, on the African continent? And
- 53:40does that differ to their attempts, for example,
- 53:42in Taiwan or in America? So far, I haven't seen,
- 53:51I haven't really been tracking a lot of Africa
- 53:53focused influence operations yet. I'm sure that
- 53:57they They do exist, but in the account pool that
- 54:03we've been tracking over the last few months,
- 54:06Africa rarely popped up. I would say that countries
- 54:13like the Philippines and Japan are a bigger topic.
- 54:20Also now, obviously, India and Pakistan, that's
- 54:22a very hot topic right now for these accounts.
- 54:28As an African, we start to see a lot of Chinese
- 54:31projects here, mostly in France. Fabulous. Great.
- 54:36Well, thank you so much, Jasper, for your time
- 54:38today. I wanted to just wrap up with a final,
- 54:43final question, if that's okay. What tools are
- 54:47absolutely vital to your deep dive research in
- 54:50general? You mentioned a couple of threads earlier,
- 54:53but which ones are absolutely vital to them?
- 54:56research that you do. We like to leave people
- 54:58with some kind of key resources to look into.
- 55:02But other than that, thank you very much for
- 55:04joining. And yeah, thank you everybody for listening.
- 55:08Yeah, if you can just throw some tools away and
- 55:11then we'll wrap. Yeah, so I really my single
- 55:15most important tool is single file. If you don't
- 55:18have access to official APIs or you don't have
- 55:22access to any scrapers, you can use single file.
- 55:26And if you have a little bit of Python or a chat
- 55:29GPT account, then you can figure it out and you
- 55:32can basically get any data and put it into a
- 55:35neat data format and then analyze it. So yeah,
- 55:39definitely just single file. I think we're going
- 55:45to have to start a little Python learning group.
- 55:49I know quite a few people know it, but I think
- 55:51we should start a little one -up in the Discord
- 55:54server and work together on that. Thanks so much
- 55:58for sharing Single File as a resource for us,
- 56:00and thanks again for joining us on this particular...
- 56:05talk. It's been super fascinating and hope to
- 56:09see you back in the Discord server again soon.
- 56:12Thanks so much, Jesper. Thank you for having
- 56:15me. Thank you, everybody. Thank you for listening
- 56:20to the stage talk. If you'd like to catch a stage
- 56:23tour live where you can ask the guest questions,
- 56:27join the Bellingcat Discord server by visiting
- 56:30www .discord .gg slash Bellingcat. The music
- 56:36you've heard is titled Dawn by Newer Self and
- 56:39is courtesy of Artlist.