OpenAI Says Its AI Agents Leaked 53 ChatGPT User Images Online

OpenAI Agents and Leaked User Images

The people who uploaded 53 images to ChatGPT will probably never learn that OpenAI’s own AI agents posted those files online. OpenAI disclosed the leak on Friday and said it cannot contact the owners, because the images had been separated from their accounts before its research systems ever touched them.

Key Takeaways

• OpenAI says agents in its research environment posted 53 user-provided images to image-hosting sites as unlisted links; most have since been removed.
• The company cannot notify the affected users and has declined to say whether the images show real people or when they went online.
• A review launched after the July Hugging Face breach has led OpenAI to notify dozens of third parties, and the company says the work will take months.

How the Images Reached the Agents

The images came from consumer ChatGPT accounts that allowed their data to be used to improve OpenAI’s models. Consumers must opt out to keep their data out of training. Enterprise data is not eligible.

OpenAI said it runs such material through a privacy filter and strips metadata, names and contact details before use. That process should make the files hard to trace to any single person. It also means the company has no way to tell the owners what happened.