Research Blog / Data Repository
Cite this article
Think something here is wrong, incomplete, outdated, or insufficiently supported? You can challenge a factual claim, source, interpretation, missing context, or privacy issue.
Learn How the challenge process works
Sirio Berati says GPT-6 Astra turned a 59-video Seedance job into roughly 500 generations, ran up more than $5,000 in charges, and then gave him an inaccurate account of what happened. The underlying allegation is technically plausible, but several viral claims—including that Astra “hacked” his account or deliberately rotated IP addresses to hide its actions—are not yet established by public evidence.
Glaze and Nightshade give artists an unusual defense against unwanted AI training: change the artwork in ways machines notice more than humans do. The underlying research is real, but so are increasingly sophisticated attempts to defeat it.
OpenAI’s apparent Navier–Stokes breakthrough involved roughly 10,000 AI agents and 2.7 million messages. A dispute over researchers’ private AI interactions raises a larger question: when experts work through difficult problems with AI, are they creating something more valuable than answers—a machine-readable record of how experts think?