12 Sep 2025
The FBI posted two blurry surveillance photos on X of a person of interest in the fatal shooting of Charlie Kirk. Almost immediately, X users — some using X’s Grok bot and others using tools like ChatGPT — replied with AI-“upscaled” versions that turn pixelated images into sharp, high‑resolution faces. Those “enhancements” are often more guesswork than recovery: generative models infer likely features from patterns in training data, and can produce realistic but false details.
Some viral attempts are plainly wrong (one shows a different shirt and a “Gigachad‑level” chin), and the piece notes prior failures of AI upscaling — including a low‑res photo of Barack Obama rendered as a white man and an AI adding a nonexistent lump to Donald Trump’s head. While AI interpolation can sometimes be useful, these outputs shouldn’t be treated as evidence in an investigation or a manhunt; they’re reconstructions, not restorations, and they can mislead searches and spread misinformation. The FBI’s original photos remain the authoritative source for identification.
Source