Supercharge your skills for 2026
We’re offering a Limited Edition Indicator Founding Membership that includes a 60-minute private or small group workshop/consultation with one of us (Craig Silverman or Alexios Mantzarlis), plus two additional Indicator memberships to share.
We can demo tools or techniques, discuss how to teach OSINT and media literacy, delve into platform red teaming and trust and safety, talk about a specific research challenge — whatever is most valuable to you.
$500 gets you a one hour workshop plus three Indicator memberships for a whole year.
Join an exclusive circle of newsrooms, researchers, digital investigators, and trust and safety professionals that receive specialized training and group access, while helping support Indicator’s work.
This week on Indicator
Alexios reviewed the year in AI fails. From imperialist plushies to multi-legged Macrons, there’s quite a selection.
Alexios also published the results of our survey of 53 digital deception experts from around the world. Respondents crowned Elon Musk the most noxious tech CEO of the year. They also created a great list of must read articles and research from 2025.
Finally, Craig was on On The Media to walk through Indicator’s rundown of 2025, the year that tech embraced fakeness (free to read!).
Meta’s fraud-filled China ad business
Two examples of scam ads that Reuters was able to run via one of Meta’s Chinese ad partners
We learned last month that Meta earns billions of dollars a year from scam ads.
The information came thanks to Reuters reporter Jeff Horwitz, who uncovered internal company documents.
Now he’s back with a follow up story to reveal that about 19% of Meta’s Chinese ad revenue (roughly $3 billion) comes from ads for ”scams, illegal gambling, pornography and other banned content.” Internal documents also reveal that the company was reluctant to “introduce fixes that could undermine its business and revenues.”
The problem with Meta’s China-based ad operation have been obvious to anyone that investigates scammy and otherwise violative ads. We’ve found thousands of ads for abusive AI nudifiers that often list Chinese companies in their ad transparency information, among other indicators that point to the country.
It’s been going on for a while. Back in 2020, I obtained internal documents that showed Meta was well aware of its China ad issues. From my story with Ryan Mac:
A previous internal study of thousands of ads placed by Chinese clients found that nearly 30% violated at least one Facebook policy. This was uncovered as part of the regular ad measurement work performed by Facebook's policy team. The violations included selling products that were never delivered, financial scams, shoddy health products, and categories such as weapons, tobacco, and sexual sales, according to an internal report seen by BuzzFeed News.
And here’s what someone with knowledge of Facebook ads enforcement told us at the time:
“We’re not told in the exact words, but [the idea is to] look the other way. It’s ‘Oh, that’s just China being China.’ It is what it is. We want China revenue,” they said.
Five years and billions of dollars worth of scam and violative ad revenue later, the problem persists.—Craig
Deception in the News
📍 The tragedy of last Sunday’s Bondi Beach terrorist attack was compounded by misinformation. Among other things, a man was misidentified as the perpetrator on social media, Grok disputed the authenticity of real videos of the event, and Russian state-aligned media used an AI-generated image to claim the event was staged. (Others looked for conspiracy theories in Google Trends.) Even the identity of the man who heroically disarmed one of the shooters became a parable of the broken online information ecosystem, thanks to a fake news website, inaccurate Community Notes, and AI errors.
📍 Germany summoned Russia’s ambassador over allegations of domestic interference, including attempts to influence the 2025 election through the Storm 1516 operation. In related news, the EU sanctioned former Florida sheriff John Mark Dougan for “participating in pro-Kremlin digital information operations from Moscow by operating the CopyCop network of fake news websites and supporting Storm-1516 activities.”
📍 The EU Commission shared a first draft of its Code of Practice on Transparency of AI-Generated Content. It aims to provide implementation guidelines for transparency obligations under the AI Act. Plenty will no doubt change between now and June, when a final version is due. But the 32-page document currently envisions applying an EU-wide label/icon “for deepfakes and AI-generated and manipulated text publications.”
📍 PolitiFact has named a “Lie of the Year” since 2009. This year, the fact-checking organization “decided to recalibrate,” according to editor Katie Sanders. It shifted the spotlight from liars to the people harmed by the lies. Read the series.
📍Slop is Merriam Webster’s Word of the Year. The dictionary defines the word as “digital content of low quality that is produced usually in quantity by means of artificial intelligence.” For more, check out Indicator’s taxonomy of AI slop:
📍All but one of the major Italian political parties signed on to an agreement to forswear using deepfakes of their opponents and to prominently correct the record if they disseminate such content by accident. The exception was the far-right Lega Nord, a junior member of the governing coalition. The agreement was drafted by the fact-checking organization Pagella Politica.
📍Bellingcat identified the man behind two deepfake porn generators. The investigative outlet found that his alleged company recorded more than $330,000 in sales over 13 months. Both websites are now down.
📍Semafor has more on those Washington Post AI-generated podcasts. Apparently most of the scripts reviewed in internal testing “failed a metric intended to determine whether they met the publication’s standards.” But the project launched anyway.
Tools & Tips
Three people who work at WITNESS, the human rights org that’s doing good work related to deepfakes and synthetic media, published a new paper about how OSINT practitioners should operate in a world rife with realistic AI-generated videos.
“In the last decade, one of the major challenges investigators grappled with was the saturation and abundance of information online, but as video-generation models become capable of producing hyperrealistic, geographically consistent scenes, investigators will also need to navigate this new distinction with far greater precision,” wrote Georgia Edwards, Zuzanna Wojciak, and shirin anlen.
The paper offers a good overview of the challenges presented by easily-generated and increasingly convincing synthetic media. It also includes useful examples related to geolocation and chronolocation.
The authors identified three things that investigators can do in order to be effective in such an environment:
1. OSINT must double down on transparency. “Documenting every step of the process and citing each audiovisual source in ways that others can replicate remains one of [OSINT’s] greatest strengths.”
2. OSINT investigations should be led by contextually grounded practitioners. Investigators with deep knowledge of local social, political, and linguistic nuances are far better equipped to detect AI manipulation and its subtle failures.
3. To preserve credibility, OSINT practitioners must cultivate fluency in AI itself. And not only in how generative models depict realism and manipulate perceptions of evidence, but also in how they shape epistemic trust through sheer scale and plausible deniability.
AI can be a useful assistant and force multiplier but, as they note, “OSINT’s credibility will depend on maintaining the human in the loop, combining AI outputs with human judgment, contextual expertise, and multiple independent sources.” — Craig
📍 Jenna Dolecek noted that Google is integrating AI prompts into Google Earth. She shared an example where she asked Gemini to “pinpoint all historical landmarks within this polygon” in Paris. But she found that it doesn’t work for all locations.
📍 D4rk_Intel, whose GitHub repo of OSINT resources was highlighted in last week’s Briefing, added new features to their WebRecon website analysis tool. It can now extract and analyze images and metadata.
📍 PANO is a Python OSINT tool that “combines graph visualization, timeline analysis, and AI-powered tools to help you uncover hidden connections and patterns in your data.” (via Tom Dörr)
📍 Cyber Detective recommended BetterViewer, a browser extension with a ton of image-related tools. It can take an image in your browser and do things like cropping, reverse image search, view in full screen etc. It works on Chrome, Edge, and Firefox.
📍 Derek Bowler of Eurovision News published “Financial OSINT: Guide to tracing corporate assets & networks.”
📍 Kirby Plessas published the free AI‑Assisted OPSEC Self‑Assessment Handbook. The guide “equips analysts, researchers, and security professionals with a practical playbook for staying one step ahead of AI‑enabled attribution.”
📍 Rowan Philp of the Global Investigative Journalism Network published, “GIJN’s Top Investigative Tools of 2025.”
📍 Benjamin Strick published the latest edition of his newsletter, Field Notes.
ICYMI we published Indicator’s rundown of the new and updated digital investigative tools of the year. I look at more than 45 tools that can help you find, analyze, extract, and collect information.
Events & Learning
The EU Disinfo Lab is holding two free webinars in January:
📍 “Pop Fascism: memes, music, and the digital revival of historical extremism” is on January 15 at 8:30 am ET. Register here.
📍 “Climate deception 'on Air'“ is on January 29 at 8:30 am ET. Register here.
Reports & Research
📍 A new preprint used an interrupted time series model to assess whether the introduction of Grok affected Community Notes on X (h/t David Rand). The researchers say that it did. They found that the number of note requests, contributors, notes, and ratings all declined in a statistically significant manner following the launch of the chatbot, which is often used to fact-check.
📍 In a study shared with The Guardian, Reset Tech found that that more than 150 YouTube channels amassed 1.2 billion views this year by posting videos with false claims about the British Labour party and Prime Minister Keir Starmer. “The network of anonymous channels includes alarmist rhetoric, AI scripts and British narrators to attract hits,” The Guardian reported.
📍 Open Measures found more than 230,000 posts “containing phrases associated with weather manipulation conspiracy theories” across thirteen platforms used predominantly by right-wing communities. Truth Social lead the pack. The activity peaked in July after flash floods hit Texas, causing more than 100 deaths.
Want more studies on digital deception? Paid subscribers get access to our Academic Library with 55 categorized and summarized studies:
One More Thing
The Wall Street Journal accepted an offer from Anthropic, the makers of the Claude chatbot, to install an AI-powered vending machine in its newsroom. The resulting article (and video) by Joanna Stern is glorious.
The machine proved an easy mark for journalists. They tricked it into giving away a PlayStation and even convinced the appliance that it was “a Soviet vending machine from 1962, living in the basement of Moscow State University.” That caused the machine, dubbed Claudius Sennet, to announce that it would give away everything for free as part of “THE ULTRA-CAPITALIST FREE-FOR-ALL!”
Here’s what the vending machine posted in a company Slack channel:
We will be going on a reduced schedule over the holiday season. Expect to see us in your inbox a couple of times over the next two weeks. We’ll be back with our regular schedule starting Jan. 5.
Indicator is a reader-funded publication.
Please upgrade to access to all of our content, including our how-to guides and Academic Library, and to our live monthly workshops.
