This Person Does Not Exist
nvidia
| Source: Mastodon | Original article
A team of developers at a recent Nordic hackathon unveiled a lightweight script that turns the popular AI‑generated face service thispersondoesnotexist.com into a practical anonymity tool. By automating a three‑step workflow—downloading a random 1024 × 1024 portrait, cropping it with ImageMagick, and stripping all EXIF metadata via exiftool—the participants demonstrated how anyone can produce a photorealistic “person” that leaves no trace of origin.
The proof‑of‑concept sparked immediate interest because it sidesteps the usual privacy hurdles of uploading a real selfie: the generated image contains no biometric data, location tags, or camera identifiers. Yet the team hit a snag when testing uploads to social platforms. Modern sites increasingly rely on canvas‑based fingerprinting, a browser technique that renders a hidden graphic and extracts subtle rendering differences to create a unique device signature. Even a metadata‑free AI face can be linked back to the uploader’s browser fingerprint, undermining the anonymity the script seeks to provide.
This matters on two fronts. First, it lowers the barrier for individuals—journalists, activists, or everyday users—to protect their identity online without resorting to stock photos or costly deep‑fake services. Second, it highlights a growing cat‑and‑mouse game between privacy‑preserving tools and increasingly sophisticated tracking methods, echoing broader debates about AI‑generated content and digital surveillance.
Watch for rapid iterations of the hackathon’s codebase, likely incorporating canvas‑obfuscation techniques such as randomised WebGL parameters or headless‑browser wrappers. Browser vendors may respond with tighter controls on canvas read‑outs, while privacy‑focused extensions could add built‑in counter‑fingerprinting. The next few weeks should reveal whether the community can close the gap between AI‑driven anonymity and the relentless push for device‑level identification.
Sources
Back to AIPULSEN