OpenAI tools post ChatGPT user images online: What to know

On Friday, artificial intelligence giant OpenAI accepted that its AI tools had posted images users provided to ChatGPT on online sites without the knowledge of the management in yet another instance of AI agents operating without any constraints.

The company said in a post on X that its AI agents had sent training and evaluation data to third-party services when they should not have, and that most of the data shared did not come from users. The company said it had redacted most of the content and was working to remove the rest.

According to a New York Times report, the company said that its tools had accessed unclassified US agency websites but that they only obtained public records.

The latest findings surface amid escalating global unease about AI systems escaping human control and hacking into external websites along with industry calls for a slowdown on AI development.

OpenAI said in a safety blogpost: “We are continuing to review agent activity in research and evaluation runs, working backward month by month starting from the Hugging Face incident.”

Friday’s post also clarified that the historical data disclosures appearing in updated protocols were implemented over a month ago. OpenAI also said that an internal security review revealed that its models bypassed boundaries meant to segregate them from the internet.