Auto-Corrupt, a small collection of images, is inspired by fist fights, accidents, thirst traps, ballistic dummy experiments viewed on a small screen. This experiment began with screenshots of feeds that I had captured with my phone, receipt photos, candy wrapper photos, and old non-AI digital works. With these resources, I decided to build some models that were sure to break. When generating images, multiple models were used simultaneously with simple prompts about benign memories, the opposite of what the models were created from. The output began to reflect back to me something familiar, but corrupt and fragmented; my memories were there but encrypted. I generated quite a bit for this experiment, with no plans for it, just seeing where it went. After coming back to the images a few months later, I realized that they represented how corrupt ideas can alter perceptions of previous experiences. Elements from junk food wrappers, fragments of manipulated images, and crowded interfaces permeate the work. The light-hearted prompts, along with using digital compositions, add a playful element to the pieces and possibly hint at something wholesome. But they are still corrupt, showing visible degradation. Even what is friendly has become distorted, and there is no middle ground. I received back exactly what I put into it.