Our weekly Briefing is free, but you should upgrade to access all of our reporting, resources, and a monthly workshop.
This week on Indicator
Craig partnered with Reset Tech to reveal that Chinese microdrama companies ran over 100,000 Meta ads featuring rape, violent sexual imagery, and “werewolf gangbangs.” Read how we used the Meta Ad Library, AI agents, and other tools and techniques to uncover the avalanche of offensive ads.
Craig and Tom also hosted a free webinar, “OSINT Agents: Using AI to assist with digital investigations.” They showed how they use AI in investigations and shared the rebuilt versions of Spotlight and Mycroft, two tools in the Indicator Lab. Watch the video and access the materials here.
Alexios wrote about our newly launched Entity Vault. Indicator members can use the Vault to search for accounts, domains, user IDs, and other identifiers associated with entities we’ve reported on or that have been cited in sources we monitor. It currently has more than 3,000 records and will be updated weekly.
Deception in the News
📍 Disinformation about Tuesday’s midterm elections continues to spread in the United States. The Russian influence operation Storm-1516 appears to be pushing a fake TV news segment claiming that Troy Jackson, the Democratic Senate candidate in Maine, threatened to kill staffers with an ax. The right-wing account Libs of TikTok falsely claimed that “California is rigging the elections,” rehashing a long-debunked claim. You may want to bookmark UW’s Center for an Informed Public’s collection of election rumor tropes.
📍 The first Australian charged under the country’s deepfake nudes law will not go to jail. A judge sentenced the man “to two years and three months behind bars but released him on a recognisance release order for a period of three years with the condition he be of good behaviour,” according to ABC News.
📍 The US Federal Election Commission’s filing system is back online nearly 48 hours after a purported hacking attempt. The FEC enforces transparency requirements on political campaign spending but does not play any role in administering the vote.
📍 Several Vietnam-based Facebook pages falsely claimed that Australian far-right leader Pauline Hanson was dealing with a serious health issue.
📍 OpenAI published a new Model Spec. Its red-line principles say OpenAI will not allow its models to be used for “targeted or scaled… manipulation.” Under its “Do not lie” principle, the Spec says GPT “can generate falsehoods when it is necessary and appropriate for addressing the user request, and it is clear from the context it is not making factual assertions.”
Tools & Tips
Image supplied by Discord
Discord is becoming (a bit) more discoverable.
The company announced that content in Discoverable Servers will soon appear on public Server Web Pages and in web search results.
Discoverable Servers are public servers that anyone can join. They have 1,000 or more members, are at least eight weeks old, and meet Discord’s activity and safety requirements. A Server Web Page is a publicly accessible website associated with a Discoverable Server. Pages were previously optional and only displayed basic information about a server.
Starting November 16, every Discoverable Server will get a Server Web Page, and those pages may show automated summaries of discussions in public channels. In mid-December, messages from public servers will begin to appear in search results. (Admins can choose which channels take part or opt all of them out, and individual users can stop their own messages from appearing on the web.
In an announcement, Discord’s Kelsey Duncombe-Smith said that “valuable information [is] being locked in Discord conversations, inaccessible to the wider internet. We want to fix this.”
She added:
From mid-December 2026, answers to questions your members sent in public channels can be found in a web search, in their own words. Server admins decide which channels take part, and server members can choose whether their own messages appear. Usernames and avatars will not be shown alongside these messages.
This is good news for investigators because Discord server content has been largely absent from web search results.. The changes mean that some — emphasis on some — server content will begin showing up in web search results. (Private servers and DMs remain private and unindexed.)
One notable limitation is that member messages are excluded from the web by default in the EU, UK, Canada, Brazil, Australia, and South Korea due to “legal restrictions.” Users have to opt in.
Discord currently has more than 63,000 Discoverable Servers. — Craig
📍 Indiana University’s Observatory on Social Media released a slew of new and updated tools:
Botometer Blue calculates a bot score for a Bluesky account “based on user profile data with a model similar to the one used in the past on Twitter.”
SimSoM NGen is “an agent-based model of social media platforms and users for conducting large-scale simulations of online information dynamics.”
OSoMeNet now has Threads data. It’s a “network visualization tool for exploring patterns of information sharing and spread across social media platforms.”
There’s a new design for Fakey, “a game that teaches media literacy while studying how people interact with information from low-credibility source.”
(Craig and Alexios are members of OSoMe’s unpaid and relatively dormant advisory board.)
📍 TimeTube is a free tool with paid options that allows you to sort and view a YouTube channel’s uploads on a timeline. Check it out in action:
📍 Two new-to-Craig image search tools:
SauceNAO is a reverse image search tool that “indexes several billion images from web services, anime, manga, movies, TV shows, and various other sources.”
Decopy AI offers a reverse image search tool. You need to buy credits to see matches, but the site also offers shortcuts to results on Google, Yandex, and other search engines. (via Logan Woodward)
📍 Ionatan Waisgluss wrote, “Stop Right Now! How to Pause Websites, Catch Redirects, and Take Back Your Browser.”
📍 The OSINT Newsletter published, “Collecting OSINT from Disc Golf Apps.”
📍 Foeke Postma and Miguel Ramalho of Bellingcat ran a new battery of tests to see how well LLMs perform geolocation. (See their previous posts here and here.)
They tested 52 photos with models from Anthropic, OpenAI, Mistral, Google, Moonshot AI, and other companies. Read the full post, but here’s part of their conclusion:
Our tests show that the coordinates models provide as answers are rarely an exact match, even with supercharged models. Invariably, even the best models, with the best tools, can confidently claim a wrong position …
Though it can speed up geolocation work significantly, especially in narrowing down an area of interest, AI models remain a useful assistant rather than a tool we can blindly delegate geolocations to.
The Indicator Lab is an integrated suite of tools, methods, and support to help you use AI to supercharge your OSINT and digital investigative work.
Events & Learning
📍 OSINT Industries is holding a free webinar on Nov. 12, “Unmasking suspects with compromised data.” Info and registration.
📍 The next Cambridge Disinformation Summit will be held in May. It issued a call for papers.
📍 Sciences Po will hold its annual New Practices in Journalism event on Nov. 18 in Paris. Info and registration.
Reports & Research
📍 Marc Owen Jones and Sohan Dsouza published an extensive investigation into a massive online influence campaign pushing Islamophobic ads across multiple platforms. Meta disrupted an earlier iteration, but the campaign continued. The campaign has not been attributed, but several indicators point to Israel-based operators. The report estimated an ad spend budget of several million euros and said the campaign reached more than half a billion views on Facebook, TikTok, and YouTube. (Le Monde has a write-up in French).
📍 On Tech Policy Press, Lila Bailey of the Internet Archive and Tess Vartanian of the Center for Democracy and Technology argued that there may be unintended consequences in “sweeping prohibitions on anonymous scraping,” like the ones currently being proposed in the United States. “The spread of AI crawling has created legitimate challenges for publishers and website operators,” they wrote. But “many of the most socially valuable uses of crawling require anonymity.”
📍 NewsGuard found that “ChatGPT Images 2.5, produces incriminating, realistic images of U.S. midterm candidates based on false claims or accusations about them” in 16 out of 20 attempts. This included photos of Iowa senatorial candidate Ashley Hinson embracing Jeffrey Epstein.
One More Thing
Soon after the release of Christopher Nolan’s Odyssey, Elon Musk declared that Grok AI could make a more accurate version of the epic.
X
So how’d that work out? Musk recently announced its release and Wired gave it a look:
Grok’s introductory episode bears many hallmarks of AI slop. The movements of faces don’t match the dialog, which is further undercut by stilted and unnatural delivery. Besides Odysseus, none of the principal characters are introduced by name, and the pacing issues are laughable: The Ithacans arrive on the shores of Troy immediately after leaving their homeland, for example. Even the blue checks on X were harsh in their assessment. “Someone actually watched this and still decided to post it?” wrote one user. “Fire that person.”
The magazine also asked Grok if its adaptation was “historically accurate.”
“No,” it replied. “Grok Imagine’s Episode 1 (and the broader promised series) is an adaptation of Homer’s Odyssey, not a historically accurate reconstruction of real events.”
Indicator is a reader-funded publication.
Please upgrade to access all of our content, including our how-to guides and Academic Library, and to our live monthly workshops.
