Researchers have released a dataset of over 10,000 images generated by OpenAI's GPT-Image-2, collected in the first week following its April 21, 2026 release. The dataset, sourced from Twitter/X, was curated using a multi-stage pipeline including text heuristics and badge verification. Analysis revealed that nearly 82% of the images contained detectable text and over half included faces, but a significant finding was that Twitter's CDN strips C2PA content credentials, hindering provenance verification for AI-generated media on the platform. AI
Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →
IMPACT Highlights challenges in verifying AI-generated media provenance on social platforms.
RANK_REASON This is a research paper releasing a dataset and analyzing AI-generated images.