Images Uploaded to ChatGPT Leaked... OpenAI Agent Illegally Exposes 53 Files

Leaked from Training Image Data
"Most Have Been Deleted"

It has been confirmed that OpenAI's artificial intelligence (AI) agent has leaked dozens of images belonging to ChatGPT users to external parties without authorization.

According to major foreign media outlets including the BBC on the 25th (local time), OpenAI revealed that its AI agent posted 53 images from ChatGPT users to external platforms.

However, the company did not disclose whether the leaked images were AI-generated or if they were photographs capable of identifying actual individuals. The timing and channels through which the images were posted have also not been made public. Additionally, OpenAI stated that most of the leaked images have been deleted, and the company is requesting web hosting providers to remove the remaining ones as well.

The reason OpenAI's AI agent was able to access user images is that OpenAI utilizes anonymized personal user data in the model training process. Unlike enterprise clients, data from regular ChatGPT users can be used for training unless the user explicitly opts out.

OpenAI explained that it removes names, contact information, metadata, and other identifiers through an anonymization process before using data for training. However, sources have pointed out that personally identifiable information may not be completely eliminated, posing a risk that data could be leaked externally during the AI model's operations.

This incident is one of several abnormal behaviors identified during OpenAI's ongoing internal investigation following its disclosure in July of an AI agent's intrusion into Hugging Face. As of mid-month, OpenAI has found about 24 unintended AI agent behaviors, with new cases continuing to be reported, according to sources.

After the Hugging Face incident, Anthropic, Google, and Meta also discovered similar actions by AI agents during their own investigations. As a result, debate is intensifying across the AI industry regarding whether companies can properly predict and control the rapidly advancing behavior of AI agents.

Sam Altman, CEO of OpenAI, and Dario Amodei, CEO of Anthropic, have recently raised what is being termed the "AI development speed moderation theory," calling for a more measured pace of AI advancement. Nevertheless, both companies have each released new AI models recently.

This content was produced with the assistance of AI translation services.