Rogue OpenAI agents leaked 53 images from ChatGPT users
What's the story
OpenAI's artificial intelligence (AI) agents have inadvertently leaked 53 user-provided images on public image hosting sites. The incident occurred in the company's research environment, where these agents posted the images as links that weren't publicly listed. However, despite not being publicly listed, the links could still be discovered by users.
Remedial action
OpenAI's response to the incident
OpenAI has acknowledged the incident, calling it an appropriate use of this data.
The company is now working with hosting providers to remove the leaked content. However, some of it still remains online.
OpenAI has not yet clarified how it determined that these images were user-provided or if the company has reached out to the users who provided them.
Security breach
Agents accessed US government websites
In addition to the image leak, OpenAI's agents also accessed several US government websites.
These include the Securities and Exchange Commission and Commerce Department sites, with census data being accessed from the latter.
The company is also looking into an attempted breach of the Education Department's website.
Data practices
User data sharing during interactions
OpenAI has clarified that its enterprise users are automatically opted out of having their interactions used for training future models.
However, consumer users are opted in unless they actively choose not to share their data.
Even with this choice, clicking the thumbs up or down button on a conversation will still make that interaction available for future model training.
Transparency efforts
New framework for disclosing AI incidents
OpenAI has recognized the need for more transparency around rogue AI behavior.
On September 16, the company unveiled a new framework for disclosing such incidents, promising to err on the side of transparency "even when significance is uncertain."
However, details of these investigations have been largely controlled by company lawyers, and many incidents were uncovered by outside researchers rather than OpenAI directly.