#AIForensics' study (07/2025) reveals how AI-generated content floods #TikTok & #Instagram, with platforms failing to adequately label synthetic material.

> #SocialMedia feeds are increasingly filled with AI-generated content that "games the algorithm" through visual plausibility & viral potential. This phenomenon, dubbed #AISlop represents sloppy, potentially deceptive synthetic content at unprecedented scale.

Interesting findings:
https://www.aiforensics.org/work/gen-ai-slop

#DSA #DigitalServicesAct

AI Generated Algorithmic Virality

This investigation systematically analyzes AI-generated content across TikTok and Instagram in three European countries. We found that 25% of TikTok's top search results contain synthetic AI imagery, with over 80% produced by "Agentic AI Accounts." Despite platforms' DSA commitments, only half of AI content receives proper labeling, creating significant risks for users.

AI Forensics

Wie künstliche Videos Social Media fluten

Studie zu “AI Slop”

Eine neue Studie zeigt: TikTok und Instagram kennzeichnen KI-generierte Inhalte nur unzureichend – und gerade TikTok wird von bizarren synthetischen Videos dominiert. Die Plattformen verstoßen damit gegen EU-Vorgaben und sorgen für eine Flut von automatisierter Desinformation.

TikTok und Instagram kommen ihren Pflichten zur Kennzeichnung von KI-generierten Videos nicht ausreichend nach. Auf TikTok dominieren zudem synthetische KI-Videos die Suchergebnisse zu populären Hashtags. Zu diesem Schluss kommt die europäische gemeinnützige Organisation AI Forensics in einer neuen Studie zur Verbreitung von KI-generierten Inhalten auf den beiden Plattformen.

Die Studie untersucht sogenannten AI Slop. Gemeint sind damit massenproduzierte, KI-generierte Inhalte, die „bizarr und sinnlos“ wirken durch ihre Kombination aus fotorealistischem Stil und unrealistischem Inhalt. AI Forensics hat analysiert, wie viele der Suchergebnissen zu den 13 populärsten Hashtags auf TikTok und Instagram AI Slop sind. Dazu betrachteten die Forscher*innen Suchergebnisse zu #Trump, #History oder #Pope mit einem geografischen Fokus auf Deutschland, Spanien und Polen.

Das Ergebnis: Sucht man nach diesen Hashtags auf TikTok, so seien von den obersten 30 Suchergebnissen 25 Prozent AI Slop. Auf Instagram ist die Anzahl deutlich geringer mit nur zwei Prozent. Von den KI-generierten Videos sind rund 80 Prozent fotorealistisch, das heißt sie sehen aus wie echte Videos. Dadurch würden sie noch trügerischer wirken.

Nicht nur Unterhaltung

Als Beispiel für AI Slop führt AI Forensics Shrimp Jesus an, eine Mischung aus Meerestieren und Jesus, die vergangenes Jahr in den Feeds von Facebook-Nutzer*innen für Irritation sorgte. Zugleich habe AI Slop auch desinformative Elemente. So identifizieren die Forscher*innen eine Kategorie von AI Slop, die sie „synthetischen Bürger-Journalismus“ nennen. Gemeint sind etwa Videos von Explosionen, die wie Amateur-Clips aussehen, aber ebenfalls generiert sind.

Ähnlich irreführend würden KI-generierte Straßeninterviews wirken, die einen vermeintlichen Regierungssturz in Iran kommentieren. Auch Videos von Politiker*innen mit falschen Aussagen hat AI Forensics laut der Studie gefunden.

Kennzeichnung reicht nicht aus

Um diesem trügerischen Effekt entgegenzuwirken, schreibt die KI-Verordnung (AI Act) der Europäischen Union Plattformen in solchen Fällen eigentlich Transparenz vor: Sie müssen KI-generierte Inhalte kennzeichnen. Dabei stützt sich die KI-Verordnung auf das Gesetz über Digitale Dienste (Digital Services Act), das vorschreibt, dass große Plattformen gegen gesellschaftliche Risiken vorgehen müssen. AI Forensics weist darauf hin, dass auch die Kennzeichnung von KI-generierten Videos dazu zähle.

Sowohl TikTok als auch Instagram kämen der Kennzeichnungspflicht nicht nach, so die Studie. TikTok gibt die Verantwortung zur Kennzeichnung von KI-generierten Inhalten an Nutzer*innen weiter und verpflichtet dazu in ihren Guidelines. Das sei nicht verlässlich genug und auch die automatisierte KI-Erkennung funktioniere nicht lückenlos, so AI Forensics.

Von den als AI Slop identifizierten Videos auf TikTok war demnach nur rund die Hälfte als KI-generiert gekennzeichnet. Auf Instagram war es noch weniger mit 23 Prozent, obwohl auch Instagram Nutzer*innen zur Kennzeichnung verpflichtet und eine automatisierte KI-Erkennung einsetzt.

Zudem seien die Kennzeichnungen schlecht sichtbar gewesen: Auf TikTok wurden sie in einer langen Liste aus Hashtags versteckt und in der Instagram-Desktop-Version wurden sie gar nicht angezeigt. netzpolitik.org hat die Kennzeichnungen überprüft und kommt zum gleichen Ergebnis.

Verbreitung läuft automatisiert

Hinter der Verbreitung des AI Slops würden fast vollständig automatisierte Accounts stecken, so AI Forensics. 80 Prozent des AI Slops auf TikTok wurde demnach von Profilen hochgeladen, die generative KI-Werkzeuge einsetzten „für automatisierte Inhaltserstellung und schnelles, repetitives Testen von Plattformalgorithmen und Publikumsinteressen“. Diese Massenproduktion von Inhalten ziele darauf ab, Algorithmen auszuspielen und die Chancen für Videos mit einer hohen Reichweite zu erhöhen. AI Forensics sagt voraus, dass dieser Prozess mit der Hilfe von KI-Agenten vollständig automatisiert werden könne.

Die Studie fordert Plattformen auf, diese Art von automatisierten Accounts zu regulieren, um eine weitere Eskalationen und „Manipulationskampagnen“ zu verhindern. Auch sollen Plattformen sich an ihre Verpflichtungen zur Kennzeichnung von KI-generierten Inhalten halten und eigene Regelungen stärker durchsetzen.

Karoline Tanck ist von Juli bis September 2025 Praktikantin bei netzpolitik.org. Sie hat Digital Media und Soziale Medien und Informationssysteme an der Leuphana Universität Lüneburg studiert. Karoline interessiert sich besonders für digitale Gewalt, den Rechtsruck und die sogenannte Künstliche Intelligenz. Kontakt: E-Mail (OpenPGP). Dieser Beitrag ist eine Übernahme von netzpolitik, gemäss Lizenz Creative Commons BY-NC-SA 4.0.

Wie künstliche Videos Social Media fluten – Beueler-Extradienst

The Conversation: Forensics tool ‘reanimates’ the ‘brains’ of AIs that fail in order to understand what went wrong. “We are computer scientists who study digital forensics. Our team at the Georgia Institute of Technology has built a system, AI Psychiatry, or AIP, that can recreate the scenario in which an AI failed in order to determine what went wrong. The system addresses the […]

https://rbfirehose.com/2025/05/02/the-conversation-forensics-tool-reanimates-the-brains-of-ais-that-fail-in-order-to-understand-what-went-wrong/

The Conversation: Forensics tool ‘reanimates’ the ‘brains’ of AIs that fail in order to understand what went wrong | ResearchBuzz: Firehose

ResearchBuzz: Firehose | Individual posts from ResearchBuzz

#FreeOur Feeds is overseen by these nine: #NabihaSyed, executive director at #Mozilla Foundation; #RobinBerjon, independent technologist; #MarkSurman, president at Mozilla Foundation; #MalloryKnodel, executive director at #SocialWebFoundation; #EliPariser, co-founder of #Avaaz; #PhilBeaudoin, entrepreneur; #MarcFaddoul, co-founder of #AIForensics; #DeeptiDoshi, co-director at #New_ Public; and #SherifElsayedAli, executive director at #FutureofTechnologyInstitute.

And the group arrives with endorsements from a set of celebrities, academics, entrepreneurs, politicians, journalists, and activists, including Wikipedia founder Jimmy Wales, Harvard Business School Professor Emerita Shoshana Zuboff, Incredible Hulk actor Mark Ruffalo, and activist and author Cory Doctorow.”

So this is the kind of grab bag technology Mozilla is backing to “secure the Internet” while slurping user data?

<https://www.theregister.com/2025/01/14/free_our_feeds_decentralized_social_media/>

Hulk smash Musk and Zuck! Actor Mark Ruffalo and non-billionaire pals back network tech underpinning Bluesky

Free Our Feeds solicits funds to foster AT Protocol that powers decentralized social media

The Register

#Facebook Is #Censoring #404Media Stories About Facebook's #Censorship

“Our findings suggest that although Meta has the tech to automatically detect #pornographic content, it does not apply it to enforce its community standards on #ads as it does for non-sponsored content,” #AIForensics said in its report. “This double standard is not a temp bug, but persisted since as early as, at least, Dec 2023.”

https://www.404media.co/facebook-is-censoring-404-media-stories-about-facebooks-censorship/

Facebook Is Censoring 404 Media Stories About Facebook's Censorship

Instagram, Facebook, and Threads are removing 404 Media stories for “nudity” as the company is paid to put ads with explicit pornography in front of its users.

404 Media

🔍 Exposing the Hidden Algorithms Shaping Our Lives

Meet AI Forensics, a European non-profit dedicated to uncovering the algorithms that influence our online experiences. In a recent interview, Katya Viadziorchyk, reveals how AI Forensics uses open-source tools to investigate tech giants and ensure algorithmic accountability.

Discover how they are safeguarding digital transparency ➡️ https://ngi.eu/news/2024/09/03/digital-detectives-ai-forensics-quest-for-algorithmic-accountability/

#AIForensics #AlgorithmicAccountability #DigitalTransparency #OpenSource

Digital detectives: AI Forensics quest for algorithmic accountability

We are delighted to announce the release of the comprehensive study report, "Benchmarking the Impact of the Next Generation Internet Initiative."

Next Generation Internet

🎉 The folks at @algorithmwatch and #AIforensics have made a formal data access request to the 🇩🇪 #DSC #Bundesnetzagentur under the new #DigitalServicesAct.

They seek to obtain #research access to internal #Microsoft data to check whether and how the company is making sure its not-so-intelligent #AI #Copilot isn't bullshitting around in dangerous ways.

Kudos to @olivermarsh and #ClaraHelming for leading this work!

https://algorithmwatch.org/en/dsa-platform-data-request-2024/
#DSA #transparency #bigtech #digitalrights #eupolicy

AlgorithmWatch and AI Forensics among the first organizations to request platform data under the DSA - AlgorithmWatch

Berlin, 15 February 2024. The EU’s Digital Services Act (DSA) is designed to give citizens new powers to protect their rights online. Today, many of its most impactful provisions are supposed to come into force: notably the creation of a Digital Services Coordinator (DSC) in every EU Member State and new opportunities for researchers to access platform data.

AlgorithmWatch
Das große #KI-Sprachmodelle unseren Meinungsbildungsprozess beeinflussen, ist ja nichts Neues. Dass sie Wahlen beeinflussen können, haben #algorithmwatch & #aiforensics vor den Landtagswahlen in Hessen und Bayern sowie den Eidgenössischen Wahlen in der Schweiz herausgearbeitet. https://algorithmwatch.org/de/bing-chat-wahlen-2023/
ChatGPT und Co: Gefährden KI-getriebene Suchmaschinen demokratische Wahlen? - AlgorithmWatch

Sich mithilfe großer KI-Sprachmodelle wie Bing Chat darüber zu informieren, wie man wählen sollte, ist eine sehr schlechte Idee. Das zeigen die Ergebnisse einer Untersuchung von AlgorithmWatch und AI Forensics zu den Landtagswahlen in Bayern und Hessen und den nationalen Wahlen in der Schweiz. Die Antworten, die große Sprachmodelle auf wichtige Fragen geben, sind zum Teil komplett falsch, zum Teil irreführend. Bing Chat und Co. können so der öffentlichen Meinungsbildung in einer Demokratie gefährlich werden.

AlgorithmWatch

Eine erschütternde Studie von @algorithmwatch ⁩ & #AIForensics zu den #Landtagswahlen in #Bayern & #Hessen: „Die Antworten, die große Sprachmodelle auf wichtige Fragen geben, sind zum Teil komplett falsch, zum Teil irreführend.“
#DSA #DSC

https://algorithmwatch.org/de/bing-chat-wahlen-2023/

ChatGPT und Co: Gefährden KI-getriebene Suchmaschinen demokratische Wahlen? - AlgorithmWatch

Sich mithilfe großer KI-Sprachmodelle wie Bing Chat darüber zu informieren, wie man wählen sollte, ist eine sehr schlechte Idee. Das zeigen die Ergebnisse einer Untersuchung von AlgorithmWatch und AI Forensics zu den Landtagswahlen in Bayern und Hessen und den nationalen Wahlen in der Schweiz. Die Antworten, die große Sprachmodelle auf wichtige Fragen geben, sind zum Teil komplett falsch, zum Teil irreführend. Bing Chat und Co. können so der öffentlichen Meinungsbildung in einer Demokratie gefährlich werden.

AlgorithmWatch
×

#FreeOur Feeds is overseen by these nine: #NabihaSyed, executive director at #Mozilla Foundation; #RobinBerjon, independent technologist; #MarkSurman, president at Mozilla Foundation; #MalloryKnodel, executive director at #SocialWebFoundation; #EliPariser, co-founder of #Avaaz; #PhilBeaudoin, entrepreneur; #MarcFaddoul, co-founder of #AIForensics; #DeeptiDoshi, co-director at #New_ Public; and #SherifElsayedAli, executive director at #FutureofTechnologyInstitute.

And the group arrives with endorsements from a set of celebrities, academics, entrepreneurs, politicians, journalists, and activists, including Wikipedia founder Jimmy Wales, Harvard Business School Professor Emerita Shoshana Zuboff, Incredible Hulk actor Mark Ruffalo, and activist and author Cory Doctorow.”

So this is the kind of grab bag technology Mozilla is backing to “secure the Internet” while slurping user data?

<https://www.theregister.com/2025/01/14/free_our_feeds_decentralized_social_media/>

#Mozilla is made up of several organizations, one of which is the #MozillaCorporation, which develops Firefox and other technologies, and another is its nonprofit and tax-exempt #MozillaFoundation, which oversees Mozilla’s corporate governance structure and sets the browser maker’s policies. Much of Mozilla’s work focused on advocating for #privacy, #inclusion, and #decentralization of #technologies, and “to create #safer, more #transparent online experiences for #everyone,” which ultimately benefit the #browser maker and its #users.

Announcing the layoffs in an email to all employees on October 30, the Mozilla Foundation’s executive director #NabihaSyed confirmed that two of the foundation’s major divisions — #Advocacy and #GlobalPrograms — are “no longer a part of our structure.””

There you go. One side of Mozilla builds the tools of the Internet, the other does… I can’t for the life of me see what. It’s the #MozillaFoundation who cant engineer the tools that are doing “deals” with others to make things safer.

The power to make change for the better of the Internet is being moved from direct impact, ie: basic technology and applied applications TO soft corporate influence through manifestos, papers, jobs and pay for influencers and collaborations with third parties…

I know what is more effective. Basic engineering and applied products because it scales.

#WhiteCollar / #Layoffs / #CorporatePolicy <https://techcrunch.com/2024/11/05/mozilla-foundation-lays-off-30-staff-drops-advocacy-division/>

Mozilla Foundation lays off 30% staff, drops advocacy division | TechCrunch

This is the second layoff at Mozilla this year, the first affecting dozens of staff on the side of the organization that builds the popular Firefox browser.

TechCrunch

Since 2022 the companies that profit from the #Internet have been shedding technical staff in the belief of the power of #AI to replace their #workforce. What are they replacing or redirecting this capacity with?

“Mozilla is eliminating its advocacy division, which fought for a free and open web: The #MozillaFoundation laid off 30 percent of its workforce and completely eliminated its advocacy and global programs divisions, #TechCrunch reports.

While Mozilla is best known for its Firefox web browser, the Mozilla Foundation — the parent of the #MozillaCorporation — describes itself as standing up “for the health of the internet.” With its advocacy and global programs divisions gone, its impact may be lessened going forward.

“Fighting for a free and open internet will always be core to our mission, and advocacy continues to be a critical tool in that work. We’re revisiting how we pursue that work, not stopping it,” #BrandonBorrman, the Mozilla Foundation’s communications chief, said in an email to The Verge. Borrman declined to confirm exactly how many people were laid off, but said it was about “30% of the current team.”

This is Mozilla’s second round of layoffs this year. In February, the Mozilla Corporation laid off around 60 workers said it would be making a “strategic correction” that would involve involve cutting back its work on a Mastodon instance. Mozilla shut down its virtual 3D platform and refocused its efforts on #Firefox and #AI. The Mozilla Foundation had around 120 employees before this more recent round of layoffs, according to TechCrunch.

“Navigating this topsy-turvy, distracting time requires laser focus — and sometimes saying goodbye to the excellent work that has gotten us this far because it won’t get us to the next peak,” wrote Syed, who previously worked as the chief executive of The Markup, an investigative news site. “Lofty goals demand hard choices.”

#WhiteCollar / #CorporateGreed <https://www.theverge.com/2024/11/5/24289124/mozilla-foundation-layoffs-advocacy-global-programs>

Mozilla is eliminating its advocacy division, which fought for a free and open web

The Mozilla Foundation, which fights for a free and open web, laid off 30 percent of its workforce in late October.

The Verge

“that would involve involve cutting back its work on a #Mastodon instance” — #NabhiaSyed, #MozillaCorporate, #Mozilla

And replacing future technical work to improve #Mastodon with this dog pile of AT Protocol concept. 👇👇👇


“Free Our Feeds solicits funds to foster AT Protocol that powers decentralized social media”

#FreeOurFeeds / #BlueSky / #Meta / #ATProtocol #ActivityPub <https://www.theregister.com/2025/01/14/free_our_feeds_decentralized_social_media/>

Hulk smash Musk and Zuck! Actor Mark Ruffalo and non-billionaire pals back network tech underpinning Bluesky

Free Our Feeds solicits funds to foster AT Protocol that powers decentralized social media

The Register

What is the AT Protocol?

Wikipedia: “The #ATProtocol (Authenticated Transfer Protocol, pronounced "@ protocol" and commonly shortened to ATProto) is a protocol and #OpenStandard for #distributed #socialNetworking services. It is under development by #BlueskySocial #PBC, a public benefit corporation originally created as an independent research group within #Twitter to investigate the possibility of #decentralizing the service.”

<https://en.m.wikipedia.org/wiki/AT_Protocol>

AT Protocol - Wikipedia

Big question, Why ATProtocol from #BlueSkySocial #PBC’s mouth instead of Mastodon and ActivityPub?:

“Why not use ActivityPub? #ActivityPub is a federated social networking technology popularized by #Mastodon.

Account #portability is a major reason why we chose to build a separate protocol. We consider portability to be crucial because it protects #users from sudden bans, server shutdowns, and policy disagreements. Our #solution for portability requires both signed data repositories and #DIDs, neither of which are easy to retrofit into ActivityPub. The migration #tools for ActivityPub are comparatively limited; they require the original server to provide a redirect and cannot migrate the user's previous data.

Another major reason is #scalability. #ActivityPub depends heavily on delivering messages between a wide network of small-to-medium sized nodes, which can cause individual #nodes to be flooded with traffic and generally struggles to provide global views of #activity.”

Short version, WE CANT CONTROL YOU.

<https://atproto.com/guides/faq>

Repository - AT Protocol

Self-authenticating storage for public account content

AT Protocol
@peterrenshaw until you can host all of bluesky in a truly federated style for less than $200 a month, account portability in BS is BS :-) #Ymmv
@roland yes that’s a limiting factor. I have no useful intuition on costs but server hosting is one I can grasp. 200 per month is a lot.
@peterrenshaw i'm no expert on decentralized computing. YES it's more difficult & flawed but at least in the case of ActivityPub & Matrix it's clearly documented and financially viable for a small org to run their own ActivityPub server and their own Matrix server unlike BS. Sure you can run your own ATProto PDS but that's it. A centralized org runs everything else. Again, not an expert and others have written better explainers and you probably knew all this anyway :-)
@peterrenshaw and who knows maybe the fine folks running Bluesky will fix ATProto (I'm unencumbered by technical knowledge :-) as always I just write Python and ruby 100 line scripts) and hosting costs will go down and it will be possible for less than $200/month to run a true peer node to the currently centralized Bluesky. That would be cool!

@roland
Inference.

the biggest lie told by startups is you use our service for free, on our servers and don’t read the terms of service.

In return we have the right to infer, sell and use data you use or view. This is the basis of todays advertising, data mining and now AI.

Having your data, on your server prevents this.

@peterrenshaw if i have read Christine Lemmer-Webber and other's posts correctly, even though you have your stuff on a 🅑S PDS, your data is still readable by the 🅑S relay.
@roland PDS?
@peterrenshaw have you not read "Christine Lemmer-Webber: How decentralized is 🅑luesky really?" <-- IMO An insightful analysis of ATProto and 🅑luesky ! https://dustycloud.org/blog/how-decentralized-is-bluesky/ PDS= Personal Data Server. https://github.com/bluesky-social/pds
How decentralized is Bluesky really? -- Dustycloud Brainstorms

@roland ah, thx Roland. I’ll be reading through this.

“> Yep! The #Firefox Browser is free.
> Super free, actually.
> No hidden costs or anything.
> You don’t pay anything to use it, and we don’t sell your personal #data.

Now it simply reads:

> Yep! The Firefox Browser is free.
> Super free, actually.
> No hidden costs or anything.
> You don’t #pay anything to use it.

Again, a pledge to not sell people's data has disappeared. #Varma insisted this is the result of the fluid definition of “sell” in the context of data sharing and privacy.

“Mozilla doesn’t sell data about you (in the way that most people think about ‘selling data’), and we don’t #buy data about you,” he said. “We changed our #language because some jurisdictions define ‘sell’ more broadly than most people would usually understand that word.”

#Mozilla / #TermsOfUse / #TOU / #AjitVarma <https://www.theregister.com/2025/03/02/mozilla_introduces_terms_of_use/>

Mozilla flamed by Firefox fans after promises to not sell their data go up in smoke

: Open source browser maker ties itself up in legalese and explanations

The Register