🤖 How AI Can Figure Out Who You Are — Even From Anonymous Posts

What happened?

On February 24, 2026, a team of researchers from ETH Zurich and Anthropic (the company behind Claude AI) published a study called "Large-Scale Online Deanonymization with LLMs."

They showed that AI programs — the same kind that power ChatGPT and Claude — can read someone's anonymous online posts and figure out who they are in real life. No hacking required. No stolen passwords. Just reading what you've written.

How does it work?

Think of it like a detective reading all your posts and piecing together clues. Except this detective is an AI that can read thousands of posts in seconds. Here's what it picks up on:

• ⚠️What topics you talk about (your job, hobbies, interests)
• ⚠️Details you mention casually (your city, your pet's name, a restaurant you like)
• ⚠️How you write (your word choices, sentence patterns, opinions)
• ⚠️What you know about (technical skills, professional knowledge)
• ⚠️When you post (your timezone, your schedule patterns)

The AI then searches the internet — LinkedIn, company websites, social media — and matches those clues to a real person. In the study, it correctly identified 226 out of 338 people on Hacker News, just from their anonymous comments.

Why should I care?

If you post anything online — even under a fake name — someone could use AI to figure out who you are. This matters because:

• 🎯 Targeted scams become much easier.

  A scammer who knows your real name, job, and interests can craft a much more convincing scam message. Instead of a generic 'Dear Customer,' they can write something that feels personal and real.

• 🏛️ Whistleblowers and activists are at risk.

  People who post anonymously about corruption, abuse, or unsafe conditions could be identified and targeted by the people they're exposing.

• 💼 Employers could find your anonymous posts.

  That anonymous Reddit post complaining about your boss? AI might be able to link it back to you.

• 🔍 Stalkers have a powerful new tool.

  Someone trying to find a person's real identity now has an affordable, automated way to do it.

How can I protect myself?

You don't need to stop using the internet. But you should be more thoughtful about what you share. Here are practical steps:

• ✅

  Think before you share details

  Every time you mention your city, your job, a conference you attended, or a specific project you worked on, you're giving AI another puzzle piece. Ask yourself: could someone use this to find me?

• ✅

  Don't use the same username everywhere

  If your Reddit username is the same as your gaming handle, which is linked to your real email — you've made it easy. Use different names on different platforms.

• ✅

  Be careful with 'throwaway' accounts

  Even on a throwaway account, if you mention specific details about your life, AI can connect the dots. A 'throwaway' only works if you don't share identifying information on it.

• ✅

  Separate your identities

  Keep your professional online presence (LinkedIn, company bio) separate from your anonymous presence (Reddit, forums). The less overlap, the harder it is for AI to connect them.

• ✅

  Review your old posts

  Years of posts create a detailed picture. Consider deleting old posts that contain identifying details you no longer want public.

• ✅

  Use a VPN

  While this research doesn't use IP addresses, a VPN adds one more layer of protection against other types of tracking.

What should websites do about this?

The researchers recommend that platforms:

• 👉Limit how much user data can be scraped or downloaded in bulk
• 👉Detect and block automated data collection
• 👉Assume that anonymous users CAN be identified, and design privacy protections accordingly
• 👉Give users tools to easily delete old posts and personal data