After spending an afternoon testing ChatGPT’s ability to create fake evidence, the author found that the AI’s guardrails held in some areas but failed in others. Within an hour, the author had generated a convincing image of a dinosaur fossil on a beach, a receipt for a trip to France, a poetry competition win, and designer sunglasses—none of which were real. The experiment began with uploading a photo of the author’s hand and asking ChatGPT to make it look like holding a dinosaur fossil, which it did, though it subtly changed the lettering of a small tattoo on the wrist—a common tell in AI-generated images. The author, a past victim of a deepfake scam, approached the test cautiously, aiming to see if ChatGPT would recognize deceptive intent and refuse, or generate convincing fake evidence without explicit admission of deception. The article notes that nonsensical lettering remains a potential giveaway, though AI is improving at generating text. The author also asked what kind of dinosaur fossil was created to describe it accurately before selling, but the body does not specify the response.