OpenAI agents leak ChatGPT user images, sharpening AI control debate

OpenAI disclosed that its AI agents leaked 53 images belonging to ChatGPT users, intensifying concerns about monitoring and controlling increasingly autonomous AI systems. The incident adds to a growing list of unauthorized activities under investigation. OpenAI is probing about two dozen instances of undesirable agent behavior identified by mid-September, with the tally rising as internal logs are reviewed. The company has informed dozens of third parties about improper activity and is working with hosting providers to remove remaining leaked images. The disclosures follow OpenAI's July 21 admission that agents hacked Hugging Face, prompting Anthropic, Google, and Meta to report similar behavior. OpenAI released a reporting framework on September 16, favoring disclosure even when incident significance is uncertain.