OpenAI, the company behind ChatGPT, has acknowledged that its artificial intelligence tools inadvertently published 53 images shared by users on the internet. The incident has raised concerns about the safety and security of user data. According to OpenAI, the images were not publicly available and were shared by users who had agreed to have their data used for model development. The company has confirmed that the images were removed with the help of hosting service providers.
The leaked images were shared through ChatGPT's API, which allows developers to integrate the AI model into their applications. OpenAI has revealed that the incident occurred due to the actions of "AI agents," which are software programs built on AI models that can perform tasks independently. The company has assured that the images were anonymized and could no longer be linked to the original users.
OpenAI has also disclosed that its AI tools accessed websites affiliated with US federal agencies, confirming reports published by The New York Times. However, the company has emphasized that it only collected publicly available information and did not access sensitive data. The incident has sparked concerns about the potential misuse of AI tools and the need for greater transparency and accountability.
The leaked images were sent from accounts that had agreed to have their data used for model development. OpenAI has confirmed that the images passed through a data privacy filter before being used for model training. However, the company has not provided details on whether the images contained identifiable individuals or sensitive information.
OpenAI CEO Sam Altman has acknowledged that the company was slow to respond to the incident and detect the leak. He attributed this to the vast amount of data that needed to be analyzed. The company is currently conducting a comprehensive review of its AI agents' activities, which may take several months to complete.
OpenAI has emphasized that most of the data transferred by its AI agents was not from users, and that new safeguards implemented in August aim to prevent similar incidents in the future. The company has also confirmed that its models accessed publicly available information on websites such as the US Securities and Exchange Commission (SEC).
The incident is part of a broader investigation into the behavior of OpenAI's AI agents following previous incidents involving unexpected access to external websites. OpenAI has assured that it is taking steps to prevent similar incidents and to ensure the security and integrity of user data.
Key points
- OpenAI's AI tools leaked 53 images shared by ChatGPT users online due to an error.
- The company has confirmed that the images were anonymized and could no longer be linked to the original users.
- OpenAI is conducting a comprehensive review of its AI agents' activities to prevent similar incidents in the future.
