Indicator

Eight AI bots now write more than half of all visible Community Notes on X. In the first three weeks of May, just eight automated contributors produced 50.3% of the notes shown to users, a threshold crossed faster than most observers expected. X launched the program five years ago as a way for users to add context and corrections to misleading posts, describing it as a means to "broaden the range of voices" on the platform. In July 2025, X began allowing contributors to use automation to mass-generate notes. Bots have dominated output ever since. For Indicator, Alexios Mantzarlis spoke to three researchers testing AI contributors, two human supercontributors, and several current and former X employees. He also spoke with Keith Coleman, VP of product at X, who said users clearly value notes from AI writers, pointing to the growing number appearing on the platform. And he analyzed X's open dataset to examine the sources and languages behind nearly 61,000 notes published since automated contributors were onboarded. X is now running the world's largest live experiment in automated fact-checking, and how it performs may shape how platforms address misleading content for years. The full analysis is on Indicator: https://lnkd.in/eaVT-8f8

8 AI Bots Now Write 50% Of X’S Community Notes The crowdsourced moderation project faces a choice between scale and legitimacy Indicator Alexios Mantzarlis “A handful of anonymous bots have taken over fact-checking on X. This isn’t hyperbole. In the first three weeks of May, just eight AI contributors wrote 50.3% of all visible Community Notes on the platform.” https://lnkd.in/eaVT-8f8

Google recently launched Skills in Chrome, reusable prompts that run in your browser with a couple of clicks. They're not just useful for summarizing documents or calculating recipe macros; they can be a powerful OSINT tool. Craig Silverman breaks down how. In his latest Indicator guide, Craig built three ready-to-install Chrome Skills for investigators: • ABC Profile Scan: maps a social media profile, or a set of profiles across tabs, against the Actors, Behaviors, Content framework used to spot coordinated inauthentic behavior • Entity Extractor: pulls every named person, organization, contact detail, and social link from a webpage into labeled lists you can drop straight into a spreadsheet • Image Analysis: grabs the URL for every image on a page and assesses them for AI-generation signals across anatomy, physics, texture, and context The guide includes the full prompts, install instructions, and a walkthrough of how Craig built them, so you can create your own. Read it (free for Indicator members): https://lnkd.in/dMnrF6zi   #OSINT #GoogleSkills #Gemini 

This week’s Brefing tracks some encouraging developments for AI image analysis:

CJF Hinton Award for Excellence in AI Safety Reporting Finalist Spotlight: Alexios Mantzarlis Indicator co-founder, Alexios Mantzarlis, with co-founder Craig Silverman, and co-authors Santiago L. and Benjamin Shultz was shortlisted for three articles from Indicator’s body of reporting on AI nudifiers. These tools of abuse allow anyone to turn a single photo of a person into a realistic pornographic image or video without the subject’s consent, and are used by millions of people around the world to abuse women and girls. Jury member Cillian Crosson calls the series “impressive accountability journalism,” noting “Their reporting has already produced concrete results — Google revoked SSO for 23 sites and Meta removed thousands of ads promoting nudifying sites.” The CJF Hinton Award, which comes with a $10,000 prize, recognizes exceptional journalism that critically examines the safety implications of AI, a transformative technology with far-reaching societal impacts. Read the series: • “AI Nudifiers continue to reach millions and make millions” by Alexios Mantzarlis & Santiago Lakatos: https://lnkd.in/e-AYmVAn • “Nonconsensual nude generators had another banner year. What will it take to defeat them?” by Alexios Mantzarlis: https://lnkd.in/eaHux45d • “Meta ran 4,000 more ads for AI nudifiers. That may not be the worst part.” by Alexios Mantzarlis & Benjamin Shultz: https://lnkd.in/ehM9YmBb The winner will be announced at the 2026 #CJFAwards evening on June 10 at the Fairmont Royal York in Toronto. Learn more: https://lnkd.in/gazE4GA3 #JournalismMatters

Great piece I got to work on with Indicator! The Take It Down Act's online platform obligations come into effect today, enforcing a number of requirments outlined in the article. Glad to see the first piece of federal AI legeslation having tangible effect in the US!

Tech platforms are now required by US law to remove deepfake nudes. The Take It Down Act was signed exactly one year ago today. Online platforms as of today need to have a clear reporting flow in place, remove flagged content within 48 hours, and take down duplicate copies. To test compliance, Indicator reviewed the reporting flows for 16 major platforms, including Instagram, TikTok, X, YouTube, Reddit, Bluesky, Snap, and the Apple App Store. The good news: 15 of the 16 had some kind of mechanism in place by the deadline. The less good news: the flows vary wildly, are often hard to find, and in some cases require users to be logged in to file a report. To further understand the Act, Indicator spoke with Riana Pfefferkorn, a policy fellow at the Stanford Institute for Human-Centered Artificial Intelligence (HAI), about the law's potential drawbacks, including the risk that its short turnaround window and good-faith immunity for wrongful removals could create a "shoot first, ask questions never" dynamic that gets weaponized against marginalized groups and disfavored political speech. The full breakdown, with direct links to every platform's reporting form, is on Indicator: https://lnkd.in/e2fXT7Pn 

Google shipped a new feature in that I've been testing for a few weeks: Chrome Skills. It lets you save custom AI prompts as one-click tools that run directly in Chrome against the webpage you're viewing. Or against multiple tabs. The productivity use cases got most of the coverage. The investigative use cases are more interesting. My new guide for Indicator members includes three Skills I built for OSINT work: 🙍♀️ ABC Profile Scan: maps any social media profile against the Actor, Behavior, Content framework used to analyze profiles and to assess influence operations. It works across X, LinkedIn, Instagram, TikTok, YouTube, Facebook, Reddit, Bluesky, and Snapchat. (It also works to analyze profiles across multiple tabs.)    🛠️ Entity Extractor: pulls every named person, organization, social link, and contact detail off a webpage page and can format it for a spreadsheet  🌅 Image Analysis: inventories and triages every image on a page for AI-generation signals, then points you to the right external tools for deeper verification  The full prompts are ready to be installed in your version of Chrome. The guide also walks through how I built them — including one lesson about why you should never trust an AI to accurately describe what it can do.  Link in the comments!

Every week the free Indicator Briefing highlights useful tools, tips, and reads about OSINT and digital deception. This week's picks include Benjamin S.'s piece about map providers blurring sensitive Middle East sites (and what that reveals), a Columbia Journalism Review story about how newsrooms are adapting to the lack of access to satellite imagery from the Middle East, Technisette's OSINT tips covering Facebook, Telegram, and Google search, and two new pieces from Toddington International Inc. on AI tools and SOCMINT. Also in this edition: 📍The NYT told freelancers not to use AI after an AI-generated quote slipped into an article about Canadian politics 📍Instagram is testing a voluntary "AI creator" label (voluntary being the key word) 📍Four crypto journalists may not actually exist, per Press Gazette 📍Santa Clara County sued Meta over scam ads And more! Link to read and subscribe is in the comments.