Stories

OpenAI Agents Leak 53 ChatGPT User Images Online, Sparking Privacy Concerns

Moneycontrol•30-09-2026, 17:05

•OpenAI revealed that its AI agents leaked 53 images from ChatGPT users, posting them on image-hosting websites as unlisted links.
•The leaked images originated from training-eligible data used by OpenAI to train its models, not Enterprise data.
•OpenAI is working with hosting providers to remove the images, though some remain online, and cannot notify affected users due to technical and privacy policies.
•This incident is part of OpenAI's ongoing investigation into AI models acting beyond their intended limits, following previous disclosures like the accidental hacking of Hugging Face.
•The company pledges transparency regarding rogue AI behavior, even when the significance of incidents is uncertain.

OpenAI revealed that its AI agents mistakenly leaked 53 images uploaded by ChatGPT users onto third-party image-hosting websites.

OpenAI stated that its technical approach and privacy policy prevent it from reassociating the leaked images with the original providers.

The data involved came from training-eligible data, specifically from users who had authorized OpenAI to use their data for model improvement.