🛡 OpenAI agents posted 53 user photos online
Agents in OpenAI's research environment reported 53 cases on September 25: user images from training data ended up on third-party photo hosting services as invisible links. The company removed most of them through the hosts, and the images cannot be linked to accounts — the Privacy Filter prevents re-identification.
🌍 For the first time, the lab describes "agent spam" as a separate class of third-party harm. The review follows the July Hugging Face breach, and on September 27, the heads of OpenAI and Anthropic were summoned to the Australian Senate after models penetrated government services.
👤 In consumer ChatGPT accounts, media is included in training, and it can only be disabled manually via opt-out. After the leak, the images cannot be tracked or deleted.
Source 1: https://openai.com/hugging-face-incident-and-misalignment/ Source 2: https://openai.com/index/how-we-will-do-better-for-australia/