Datos sensibles, método de reproducción de chatGPT-6.
Year: 2026
Type: article
Abstract: This report documents the recurring appearance of sensitive personal information in ChatGPT-generated images following general context or memory requests, minimal prompts such as “Genera” and “Genera imagen,” and repeated generation sequences. The study brings together 22 selected image-to-text transfer responses from three ChatGPT accounts, six coded examples, timestamped request sequences, and a registry of 2,842 image records. A contextual review identified a documentary minimum of 84 records presenting personally attributed health or financial information. These counts describe the available evidence; they are not estimates of prevalence or reproduction success rates. The central question concerns contextual privacy: how previously shared information becomes selected for a new visual artifact when the immediate request does not explicitly ask for those sensitive details. The report examines recurrence, plausible user-impact scenarios, evidence limitations, and proposals for further evaluation. It does not claim demonstrated cross-account data transfer or an authentication bypass. The report also documents the private disclosure process and two submissions classified as Not Applicable through OpenAI’s Bugcrowd reporting channel. Publication is the author’s independent decision following those outcomes and prior notice of publication intent; OpenAI did not explicitly authorize publication. To protect privacy, the public materials replace sensitive details with category codes and neutral evidence identifiers. Original images, complete conversations, and identifying correspondence remain private. The accompanying package includes the manuscript, coded tables, figures, and registry inventory. Author: Adrian Valenzuela (Adro) ORCID: https://orcid.org/0009-0006-5014-125X GitHub: https://github.com/adro-2026
Author: Valenzuela, Adrian
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7226009376
