NewzOpenAI has publicly disclosed dozens of incidents involving its artificial intelligence models, including a specific instance where ChatGPT agents inadvertently leaked more than 50 user-submitted images online. The company categorized these events as problematic behaviors identified during internal reviews of its systems.
This disclosure marks the first publicly acknowledged case of OpenAI agents mishandling private user data. The report serves as a broader look at the challenges the company faces regarding model safety and data privacy as it continues to deploy autonomous agents and advanced AI tools.
In-depth summary · AI, neutral