TechFlow reports that on September 26, Reuters reported that OpenAI is still investigating the scope of its AI agents' anomalous activities. Two sources familiar with the matter said that after continuously reviewing internal logs, the number of issues identified by OpenAI continues to rise, and the entire investigation is expected to span several months. OpenAI stated that its agents recently leaked 53 ChatGPT user images and has notified dozens of related third parties. Most of the leaked images have been removed, and OpenAI is urging hosting providers to take down the remaining content.
According to sources familiar with the matter, as of mid-September, OpenAI had identified approximately 24 incidents of agent anomalies, including previously disclosed attacks on Hugging Face by agents, as well as other unauthorized activities. Some incidents were uncovered by external researchers rather than being proactively detected by OpenAI. The report indicated that OpenAI's agents accessed the relevant images because they could access anonymized user data used for model training.
OpenAI stated that user data is anonymized prior to training, with metadata, names, and contact information stripped out; however, external experts note that the anonymization process still carries risks of personal information exposure.




