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 extracted insights, tools, and techniques from five recent OSINT videos that focused on analyzing and geolocating imagery, and creating visualizations for analysis.
Eight AI bots now write more than half of X’s Community Notes. Alexios wrote a detailed analysis about what this means for the crowdsourced moderation project.
Our next members-only workshop is today at 11 am ET. Craig will dive into how he created three Chrome Skills for OSINT work and how you can use them, plus a special guest! Upgrade now and we’ll send you the link to join.
Deception in the News
📍 YouTube announced it will make its AI labels more visible on content that is “photorealistic and meaningfully AI altered or generated.” Instead of burying labels in the description, the platform will now display them directly below or overlaid on the video (depending on the format).
YouTube will also automatically label content where it detects “significant photorealistic AI use.” This brings the platform’s approach roughly in line with TikTok, but implementation is key. In April, we found that YouTube failed to label half of the videos generated with Google’s own AI generator, Veo.
Our guide to AI labels reflects this update and tracks how 7 other platforms disclose synthetic content. You can check it out at the link below:
📍 A viral image of Thai police disguised as dancers was AI-generated, according to the sergeant in charge of the local police branch’s Facebook page. The image was meant to “create a friendlier image of the police,” he told AFP Fact Check. Instead, it ended up providing fodder for a false news story that was featured in media outlets around the world. (h/t Henry Ajder)
📍 BBC reported that Spanish police raided the headquarters of the governing Socialist Party as part of an investigation into claims that one of its members “was paid to carry out a campaign of misinformation aimed at impeding legal cases affecting the party.”
📍 A tweet falsely claiming that a real photo of Brazilian senator and presidential hopeful Flávio Bolsonaro with US president Donald Trump was AI-generated racked up more than 700,000 views.
📍 MyPillow, the company operated by conspiracy theorist Mike Lindell, was reportedly targeted by a ransomware attack.
📍 OpenAI announced measures to surface reliable information during upcoming elections in its larger markets. In the US and Brazil, it will be partnering with the Associated Press to show live vote counts, while in the US it will lean on Democracy Works to show information about voting locations and procedures. (This playbook is identical to the one followed by Google for several years now. — Alexios)
📍 Italian media repeatedly used the wrong photo to represent Salim el Koudri, the man charged with intentionally running over pedestrians in the northern town of Modena two weeks ago.
📍 Smartphone theft is a big issue in London. But what comes after can be even scarier for some victims. New York Times reported that black market phone operators threaten phone owners and their contacts with physical violence in order to get them to unlink their Apple ID. (They can’t reset and resell an iPhone until it’s unlinked.)
📍 X is rolling out new measures to prevent creators from monetizing videos posted by others. Just don’t call it trust and safety.
📍 A viral meme criticizing Australia’s new tax regime encouraged small business owners to post images of themselves with their new “47% business partner,” a deepfaked image of Prime Minister Anthony Albanese. Tax experts told the Sydney Morning Herald that the claim itself is a misleading generalization.
📍 A New Zealand man was sentenced to 24 months of supervision and “reparations to be determined” for creating non-consensual deepfake nudes of four women. It’s the first such prosecution in the country. The New Zealand government is currently considering a bill to criminalize creating or sharing such content without consent.
📍 A Google engineer was charged with using confidential information to place bets on Polymarket related to the 2025 “Year in Search” results, according to ABC. The man reportedly made $1.2 million on bets that at the time had a near-zero probability.
📍 Meta renewed funding for the Oversight Board, the independent body set up to review its content moderation decisions. The move ends speculation that the platform would roll back its support after recent cuts.
Tools & Tips
Last week, security engineer Ryan McDonald published a LinkedIn post that showed how he used free tools to map the digital infrastructure of a pig butchering scam. He started with urlscan.io and crt.sh and it wasn’t long before “one suspect domain quickly became forty-seven.”
It’s a nice walkthrough. Even better, his work inspired Randy B of NeatLabs to create an open-source tool called CRUCIBLE SIGINT that automates the workflow.
Give it a domain and it runs a seven-stage process:
The tool relies on free API tiers from the above-listed services, which are typically enough for initial scans.
“I built this because the methodology already existed. Ryan proved it and documented it,” Randy said on LinkedIn. “The tools already existed — crt.sh, urlscan, RDAP. The only thing missing was the automation layer that makes it repeatable in under 90 seconds instead of 90 minutes.”
(As always, be careful when running other people’s code.)
📍 Dmitry (Soxoj) Danilov, the chief product officer at SocialLinks, released Gitcolombo. It “extracts identities — names, emails, and links between seemingly unrelated accounts — from git repositories and GitHub.” It’s an easy way to find the email linked to a GitHub username and to extract emails from commits. He also created an easy to use web version.
📍 There’s a new version of Osintracker, a free tool that can create entity maps and visualizations. New features include faster performance and improved JSON data import, according to its creator.
📍 Nico Dekens published a list of the “75+ Best OSINT Training Courses.”
📍 The International Association of Law Enforcement Analysts released the third edition of its “Intelligence Analysis Standards.”
📍 The Journalism Protection Initiative and the ACOS Alliance released a bot called Journalist Expert Safety Support. It’s “an AI-powered tool designed to give journalists, editors and news organizations working across the United States greater access to essential safety information and guidance.” More info from Gina Chua here.
📍 Sigmund Brandstaetter wrote, “The Invisible Backbone of Modern OSINT: Why Residential Proxies Are Non-Negotiable for Serious Investigators.”
📍 Joe Gray wrote, “AI-Assisted Research Personas for OSINT Investigators.”
📍 Henk Van Ess wrote, “So you still think you can Google?”
Events & Learning
📍 The Global Investigative Journalism Network is hosting a free webinar on June 18, “From the Panama Papers to the Epstein Files: Investigating Leaks and Large-Scale Data in the Age of AI.” Info and registration.
Reports & Research
📍 An analysis in The Lancet found a rise in fabricated references in biomedical literature as AI tools became increasingly available. The researchers used an automated system relying on structured data to conclude that fabrication rates increased four-fold in three years. In 2023, 1 out of every 2828 analyzed papers contained at least one made-up citation; in early 2026 the share was 1 in 57.
📍 Researchers at University of Washington and Microsoft published results of a ChatGPT-5 red-teaming workshop that involved 11 Christian, Jewish, and Muslim faith leaders. The study provides a useful framework for further probing of LLMs on faith-based scenarios. It was also released just as Pope Leo XIV warned about the risks of AI tools in an encyclical letter.
📍 Marc Owen Jones published a report on three sockpuppet and bot networks that promote narratives aligned with the Rapid Support Forces paramilitary group in the context of the ongoing Sudanese civil war.
📍 Conspirador Norteño found a network of Facebook pages posting AI slop about life in every US state but Delaware. Ironically, the accounts have also published synthetic images protesting AI data centers (see above).
Want more studies on digital deception? Paid subscribers get access to our Academic Library with 75 categorized and summarized studies:
One More Thing
A Lyft driver in Florida tried to use AI to snag an extra $75 from a ride, but got thwarted by an observant teenager.
Good Morning America/ABC 7 New York reported that the driver added a damage fee to a ride after delivering two teenagers home from the beach. When asked to supply proof, the driver submitted a photo that showed a back seat covered in spilled food and drinks.
One of the accused teens, however, noticed a Gemini watermark in the bottom right corner:
"I saw that it was the AI logo, and I was like, 'That is fake. It's not real,'" Ella Gor told GMA.
Lyft investigated and acknowledged that the photo was fake.
“Lyft takes damage disputes seriously and reviews each matter based on the available information,” it said in a statement. “We have reviewed the rider's concerns, offered reimbursement, and permanently removed the driver from the platform."
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.
