While skimming through tens of Doppelgänger ads recently approved by Meta, for a total reach over 200k users in France, Germany, Italy, and Poland over the past two weeks, I stumbled upon an interesting slip-up in one of their ads 👇
In the foreground, we see the image they likely wanted to use, sent by an "Алина Малинина" via their VK Teams . On the left, there are distinct discussion channels for the countries targeted by the Doppelgänger operations: Ukraine, Germany, France, Israel, US, Italy, and Poland.
Instead of uploading the desired image, they accidentally uploaded a screenshot of it, took last Friday, revealing their VK Teams interface in the background. After a loooooong process of enhancing the compressed image, I managed to decipher it 👀
With
@aiforensics_org
we published multiple reports revealing Meta's systemic failure in ads moderation and the massive scale of the operation. Also, big thanks to
@antibot4navalny
for their amazing work monitoring Doppelgänger bots on X and helping analyze these findings!
The French discussion channel, “S-ф”, was open, revealing a conversation on the right. Алина and colleagues shared a link to a fake
@le_Parisien
article and messages that seem to read: “Changed the text of the task to 'You […] that Ukraine can win […] the war? I don’t'.”
Their choice of logos is quite telling: a beer for Germany, cheese for France pizza for Italy and Winnie-the-Pooh for their channel dedicated to websites and animations🧸.
It shows multiple targeted countries, organized teamwork, and a multi-faceted approach involving ads on Meta (which has neglected to address systemically) and spoofed news outlets shared on X.
Without overstating the findings (and acknowledging the possibility of a trap), this screenshot, extracted from a Doppelgänger Meta ad published last Friday, supports our current understanding of their operations.
Over the past week, Meta approved and delivered pro-Russian propaganda ads to more than 413,000 accounts. No systemic measures appear to have been implemented to safeguard the integrity of the upcoming EU elections !
#breaktheglass
Meta approved and delivered to thousands of French accounts Doppelgänger ads calling to vote for far-right candidates on Sunday, with xenophobic and anti-EU narratives. Playing hide-and-seek with threat actors instead of a systematic fix unsurprisingly fails.
After publishing its Adversarial Threat Report on Wednesday, Meta approved new pro-Russian propaganda ads to run on Thursday, which linked to spoofed domains impersonating news media. No systemic measures appear to have been implemented to moderate political advertisements
Wanna bypass Meta's 'state-of-the-art' moderation systems? Use emojis! Last week, Meta reviewed, approved, and delivered doppelgänger propaganda ads with a total reach exceeding 200k in France alone!
🚨Influential, yet unexplored. I'm thrilled to share my preprint exploring Amazon's recommender systems! Mapping the book recommendation landscape, we find tight communities, making it tricky for users to break free.
🚨 After our recent study revealed Meta's flawed moderation of political ads, enabling massive foreign interference weeks ahead of the elections, the EU Commission is launching an investigation into potential Digital Services Act violations.
Today we open cases against
#Meta
for suspected breach of
#DSA
obligations to protect integrity of elections:
▪️Inadequate ad moderation exploited for foreign interference & scams
▪️Inadequate data access to monitor elections
▪️Non-compliant tool for flagging illegal content
With the French snap elections in less than two weeks, Meta seems not to have implemented any systemic fixes. They recently approved Doppelgänger ads congratulating the far-right for their EU election results, framing it as revenge for Macron's support of Ukraine.
With the EU elections just weeks out, Meta still neglects to enforce its political ads policy, allowing covert political ops, including massive pro-Russian propaganda. A long thread 🧵
"After all, sanctions, inflation and wars are no reason to give up on your appearance." This new Doppelgänger clone is being promoted by ads on Meta and has been shown at least 60k times since Friday to at least 50k unique French accounts so far.
🧵 Women's magazine on well-being has just got a Doppelgänger clone.
"At times of sanctions, inflation and wars, here are beauty tips with inexpensive home-made products."
Who said that Kremlin is most stupid and straightforward in its propaganda? Your feedback was heard. 1/
Cette campagne s'inscrit dans la continuité des efforts cross-plateform de Doppelganger. Le mois dernier, Meta avait approuvé des publicités similaires.
Le collectif
@antibot4navalny
a identifié une nouvelle campagne de
#d
ésinformation anti-ukrainienne. Les Bots du Kremlin ont diffusé 120 000 messages avec fausses citations de célébrités dont Alain
#Delon
, Lea
#Seydoux
, Luc
#Besson
, Elizabeth
#Debicki
et Michelle
#Mercier
Following psychologies[.]top's tips on looking like a princess despite wars and inflation, Doppelgänger has now spoofed closermag[.]eu. Thanks to Meta's advertising systems, they delivered to thousands of users 'life advice' that 'even Mr. Great Depression Macron can't ruin'
Misleading search results, user confinement, and the recommendation of sexual content, with
@aiforensics_org
and
@CheckFirstHQ
we analyzed Amazon's bookstore algorithms within the DSA framework. Check it out!
Less than 4 weeks to the EU elections and
@Meta
is still letting foreign interference slide on its ads platform! If we can detect these campaigns, so should Meta. Time to step up and act on the commitment to "preserve the integrity of these elections on our platforms."
Now that the Poland logo channel has been identified (thanks
@alex___liberty
), one mystery remains: Why would Russian propagandists mention John Galt? Any clues?
🔍 We've identified a large community of books co-recommended by Amazon promoting climate denialism, COVID-19 conspiracy theories, and conservative views on gender.
Explore Amazon's recommendation landscape for yourself at
Besides propaganda suggesting the EU might send troops to Ukraine to boost U.S. arms sales, Meta is also allowing pro-Russian ads claiming
@EmmanuelMacron
would rather spend billions on Ukraine than help single mothers
Thanks to the DSA, we now have access to all ads targeting EU countries. As analyzed by
@CheckFirstHQ
with
@mozilla
, Meta's Ad library is quite complete, compared to X's.
@Meta
@Meta
considers the content of this ad to be against their Advertising Standards, yet this pro-Kremlin propaganda ad has been shown to 252,528 accounts, and is still running.
@SourcesOuvertes
@Colon_David
Ce compte a également massivement promu cet article en mai dernier via X Ads, exposant 81k utilisateurs uniques (109k vues)
I trained a model to classify whether an ad should be declared as political under Meta's political ads policy. Then applied it over 30 million ads published in January-February 2024 in 16 EU countries.
As the EU elections are coming up this Sunday, let's take a closer look at the French political discussion on X. By pulling together reposts from politicians and activists, we see a similar structure as the one reported in 2023 by
@ISCPIF
Hey
@Meta
, what measures are you implementing to prevent foreign interference ? There appears to be room for improvement in your detection of coordinated inauthentic behaviors.
#DSA
In the final days of the EU campaign, Meta kept failing to detect undeclared political ads during their initial review, allowing the spread of propaganda to EU citizens.
Hey
@Meta
, why don't you apply your moderation decisions systematically across duplicated ads ? Your lax enforcement of your own advertisement policies allows pro-Kremlin propaganda ads to be run weeks ahead of the EU elections
Meta is aware of these scams, widely reported in the news, and thoroughly investigated by
@CheckFirstHQ
in June 2023, who communicated their findings directly to Meta's security teams.
@Meta
Beyond disinformation and pro-Russian propaganda, Meta now approves vote instructions against
@EmmanuelMacron
for the June 9 EU elections without political disclaimer
📢 New Preprint exploring Members of Parliament reach in algorithmic timelines using engagement predictive models! Do right-wing parties benefit more from engagement-based recommender than left-wing parties ? It depends on how you measure amplification!
Hey
@Meta
, why did you just approve these pro-Kremlin propaganda ads to be run without a political disclaimer ? Are your systems fooled by a simple nickname to designate
@EmmanuelMacron
?
ii) Doppelgänger! Since August 2023, we detect nearly 4k ads relaying pro-Russian propaganda, with a total reach over 38 million accounts in France and Germany.
@NW28600848
They systemically target 'woman' with different age ranges but always 32 and beyond. 'Women between 55-64' are the most reached, with a total of over 11k.
In posts about Israel-Palestine: the left condemns the action of Israeli military and demands adherence to international law, while the presidential majority and (far)-right-leaning groups echo messages from the Israeli embassy and criticize pro-Palestine protests at SciencesPo.
Amazon and Fnac algorithms tend to deliver like-minded content to users. Rather than mere technical tweaks, we argue, with
@BuseCett
, that systemic changes are needed to mitigate their impact on civic discourse.
Thrilled to announce the publication of my latest research in the Journal of Computational Social Science. I explore how engagement-based recommender systems affect the diversity of content displayed to users on social media.
The narratives and volume of the campaign evolve with (inter)national events. The propagandists weaponize news events, such as the farmers protests and military aid packages, to undermine support for Ukraine and institutional support in the EU.
These coordinated campaigns are quite easy to spot, the proof, I did it with only a fraction of the information available to Meta. The pages are used a single time to publish an advertisement and have common name generation patterns.
Meta's moderation of political ads performs the best in Ireland, the only English-speaking country in the study, catching up to 22% of undeclared political ads, but only 3.9% in Hungary.
Less than 20% of such pro-Russian propaganda ads were moderated by Meta as political, despite falling fully under its policy, and only after being displayed by Meta between 2.6 and 3.6 million times to users.
When looking for coordinated campaigns (different pages publishing highly similar ads), we uncover i) a massive investment scam network, active in 10/16 countries in Jan-Feb 24, totaling a reach of 128 million accounts.
Looking forward to presenting my work on algorithmic amplification, engagement maximization, and social media audit methodologies at
#COMPLEXNETWORKS2023
📢 New Preprint exploring Members of Parliament reach in algorithmic timelines using engagement predictive models! Do right-wing parties benefit more from engagement-based recommender than left-wing parties ? It depends on how you measure amplification!
In France, on average, the pro-Russian ads reached 140k accounts daily. Days after
@SecBlinken
announces a $1B aid package for Ukraine, ads suggesting that the US. instigated the conflict to bolster arms sales reached 132k accounts in France in few days
I trained machine learning models, on behavioral data, to predict user engagement with tweets. This reveals how the pursuit of engagement maximization skews content visibility, favoring posts similar to previously engaged content while downplaying alternative perspectives.