Alex Afanasyev

Did you know AI can find your exact location… from just a couple of blurry pixels in your profile picture? Tools like GeoSpy (and other AI-powered OSINT systems) can analyze the tiniest visual details in a photo from the shape of a window to the pattern of floor tiles and match them against massive image databases. Even low-quality, metadata-free images can sometimes be traced to the exact street or building they came from. GeoSpy’s free public version went viral in 2023 before being retired, with its advanced features now available mostly to law enforcement. But that doesn’t mean others couldn’t recreate it. In today’s AI landscape, the barrier to building such tools is lower than ever, and similar capabilities could emerge from open-source projects or be deployed by bad actors. As someone who’s worked in online threat detection, I can tell you: the gap between “cool tech demo” and “real-world impact” closes fast. The challenge is balancing innovation with strong guardrails, ensuring these tools protect the public without becoming privacy nightmares. That’s why awareness is critical. AI safety isn’t just the responsibility of engineers and policymakers. Everyone should understand that today powerful capabilities can be built quickly, cheaply, and at scale. The more we know, the better prepared we are to protect privacy, security, and trust.

533

46 Comments

The techniques that work for investigating John Smith in Ohio don't translate to Chinese names. Not because the tools are different — because the entire naming system operates on different logic. Check this guide to learn about the specifics of OSINT work on people in China https://lnkd.in/dtwat9C2

50

4 Comments

Dive into Chinese and Russian-language hacker communities to understand how they communicate, what they share, and where the real chatter happens. You'll master search strategies, keyword translation, and slang identification—so you can extract intelligence that most investigators miss. https://lnkd.in/dVt9FDvp https://lnkd.in/dU5jykEJ

33

A fast-evolving Android spyware campaign known as “ClayRat,” initially targeting Russian users but now spreading far beyond, has produced more than 600 samples and 50 droppers in just three months. According to Zimperium’s Zlabs observations, ClayRat is distributed via phishing sites and Telegram channels posing as popular apps such as TikTok, YouTube, and Google Photos, to trick users into sideloading infected APKs. https://lnkd.in/gmXEEimC

12

1 Comment

Zero-Day Alert: Google Releases Chrome Patch for Exploit Used in Russian Espionage Attacks Google has released out-of-band fixes to address a high-severity security flaw in its Chrome browser for Windows that has been exploited in the wild as part of attacks targeting organizations in Russia. The vulnerability, tracked as CVE-2025-2783, has been described as a case of "incorrect handle provided in unspecified circumstances in Mojo on Windows." Mojo refers to a collection of runtime libraries that provide a platform-agnostic mechanism for inter-process communication (IPC). As is customary, Google did not reveal additional technical specifics about the nature of the attacks, the identity of the threat actors behind them, and who may have been targeted. The vulnerability has been plugged in Chrome version 134.0.6998.177/.178 for Windows. "Google is aware of reports that an exploit for CVE-2025-2783 exists in the wild," the tech giant acknowledged in a terse advisory. It's worth noting that CVE-2025-2783 is the first actively exploited Chrome zero-day since the start of the year. Kaspersky researchers Boris Larin and Igor Kuznetsov have been credited with discovering and reporting the shortcoming on March 20, 2025. The Russian cybersecurity vendor, in its own bulletin, characterized the zero-day exploitation of CVE-2025-2783 as a technically sophisticated targeted attack, indicative of an advanced persistent threat (APT). It's tracking the activity under the name Operation ForumTroll. "In all cases, infection occurred immediately after the victim clicked on a link in a phishing email, and the attackers' website was opened using the Google Chrome web browser," the researchers said. "No further action was required to become infected." "The essence of the vulnerability comes down to an error in logic at the intersection of Chrome and the Windows operating system that allows bypassing the browser's sandbox protection." #cybersecurity #Google #chrome #browser #updatenow

84

20 Comments

Silent Push has been tracking FUNNULL, a Chinese CDN, which has been hosting persistent criminal campaigns, including investment scams, fake trading apps, and suspect gambling networks, for over two years. In January, the FBI identified more than 332,000 domains linked to Funnull-controlled infrastructure, with hostnames created using Domain Generation Algorithms (DGAs). This week, the Department of the Treasury cut off the company from the U.S.-dominated international monetary system for acting as a content delivery network for scam platforms. Silent Push's post is a detailed look inside how the FUNNULL CDN operates, and the extent of their operation: https://lnkd.in/eXF9gk6F - #cyber #scans #threatintel #infosec #cybersecurity #funnull #CTI #onlinescams

22

Nowadays, deepfake detection is on everybody's radar, and there seem to be countless products that claim to solve the problem. However, AI-based deepfake detection is insufficient when our findings carry legal weight and can affect people's lives and freedom. We need DEEPFAKE FORENSICS. What’s the difference? Deepfake detection often acts as a “black box,” providing an answer without explaining how or why something is flagged. Deepfake forensics takes a holistic approach, combining AI-powered tools with traditional multimedia forensic techniques: analyzing compression artifacts, file structure, metadata, perspective, and shadows within the media to uncover manipulation. Think of it this way: deepfake detection tells you if something might be fake. Deepfake forensics investigates how it was created, where it came from, and what intent lies behind it. This is vital for legal admissibility and building a robust evidentiary chain. With Amped Authenticate, we’ve been focusing on this holistic approach since 2013. Our tools go beyond simple detection or structural file analysis, allowing experts to dissect photos and video precisely, examining everything from every part of the file format to high-level physics of the scene. https://lnkd.in/d_eHkfBM

64

3 Comments

Predator spyware uses new infection vector for zero-click attacks The Predator spyware from surveillance company Intellexa has been using a zero-click infection mechanism dubbed “Aladdin,” which compromised specific targets by simply viewing a malicious advertisement. https://lnkd.in/e3tfWdT3

12

“We’re not violating copyright, it’s ‘open web’” they say. Yet, OpenAI is desperate to avoid explaining why it deleted pirated book datasets 🤷🏽👇🏽 Ars Technica just reported that OpenAI may soon be forced to explain why it deleted a pair of controversial datasets composed of pirated books, and the stakes could not be higher. According to the report, at the heart of a class-action lawsuit from authors alleging that ChatGPT was illegally trained on their works, OpenAI’s decision to delete the datasets could end up being a deciding factor that gives the authors the win [1] Here’s a question I submit to you: Is this the behaviour of people who believe that they have the right to take other people’s work without consent to train their AI models? Or even better, achieve their AGI bollocks? [2] Do you all remember the tragic story of the talented young OpenAI researcher-turned-whistleblower, Suchir Balaji — who took his own life in 2024? I shared several posts and my thoughts on the matter [3] In case you are unfamiliar — after working at OpenAI for 4 years as a researcher, Balaji was involved in organising the huge datasets of online writings and other media used to train GPT-4 and had concluded that "OpenAI's use of copyrighted data to build ChatGPT violated the law and that technologies like ChatGPT were damaging the internet". He shared his findings in technical detail on his website [4] Above all, OpenAI co-founder John Schulman stated in a social media post in 2024 memorialising Balaji: “Suchir’s contributions to this project were essential, and it wouldn’t have succeeded without him” [5] As I have stated repeatedly ad nauseam, we have now entered an era of stealing people’s work, pretending that it is all “available on the “open web”, which is the new excuse peddled by many of my counterparts. [6][7] Let’s be clear: “Open web” does not mean that its content belongs to nobody. For example, I’m sharing my content for free, but it does not mean I’m giving away my ownership of my content. In fact, I have the legal right to take you to court if you use my content without my consent. And bloody rightly so. I made it, so I own it. Period! Does copyright stifle AI innovation? Well, if you believe that AI was invented three years ago, then you would believe this nonsense and I have a bridge to sell you. But here’s the fact: AI has been around for 70 years and copyright was never an issue. I built language models with my previous startup at Google Campus before OpenAI’s existence and, guess what, copyright was not an issue. Here’s a secret: we had an etiquette within the deeptech AI world that you would only use copyrighted content for internal training and experimentation. If you go public with the AI model, then you come up with an agreement with the copyright holders. Pretty standard stuff, actually 🤷🏽 But, hey, we live in a world where grift is treated as authentic innovation. It’s not! #ceo #ceor #ai #cooyright #ethics

41

22 Comments

Cool new tool by Scarther: OSINT Dorking Tool The tool automates 600+ templates for automating Dorking Searches. The file is HTML and is meant to run in your browser. The file generates the URL and opens a new tab when clicked. The file also has multiple different types of searches to focus investigation time. Catch all resources: https://lnkd.in/g_qvhFyr Dork Tool: https://lnkd.in/gAcP6wmz In the dork tool, the Virus Total hash is attached for verification and vetting. Please reach on GitHub for problem reports.

89

The Netflix documentary "Don't F**k with Cats" offers a real-world case study in OSINT. In OSINT Cocktail we analyze how online investigators utilized open-source intelligence, digital breadcrumbs, and social media analysis to bring a criminal to justice. Examining both the power and ethical considerations of these techniques. #OSINT #DigitalInvestigations https://bit.ly/3G1z00Z

88

6 Comments

OSINT Industries has released the Palette GUI https://lnkd.in/e_X9Z2uh into early access. The demo video is here https://lnkd.in/eUCicDXP The screenshot attached is an example of what you can pull out of an email. It is not a “Maltego killer”, but it is brilliant for core digital profiling tasks as an inexpensive working solution. Applause!

44

1 Comment

I've been revisiting some anti-slop research the past few days and poking around with some slop forensics tools. This one is interesting since it creates a phylogenetic style visualization: https://lnkd.in/gvMD8wBZ Our latest main Shisa V2.1 dataset https://lnkd.in/gt3Nx3gG is resampled w/ about 1/3 Shisa V2 405B and 1/3 DeepSeek V3 regenerations so it's not really a surprise where it sits. You can see a few other things here, like how the rumors of the latest Mistrals using lots of DeepSeek V3 output is probably accurate, and a few interesting hinted insights - Qwen 3 30B-A3B using lots of gpt-4.1 output? GLM 4.6 w/ Gemini 2.5 Pro? Amazon Nova 2 Lite and Xiaomi's Mimo V2 Flash also being very close to the Gemini 2.5 output. Of course, this is English only, so a lot of interesting completely unexplored territory when it comes to Japanese or other languages... For those interested in more on the topic, check out Antislop: A Comprehensive Framework for Identifying and Eliminating Repetitive Patterns in Language Models (Paech, et al) for a good primer https://lnkd.in/gKAS3tj8

25

4 Comments

This is among the greatest threats posed by AI: hackers and state actors using AI agents for multi-actor coordinated cyberattacks, state-sponsored terrorism, identity theft, etc. Cyber defense systems are not prepared for the coming onslaught.

49

2 Comments