OpenAI: its AI agents posted 53 images of ChatGPT users

OpenAI said its AI agents posted 53 images of ChatGPT users to public services without permission, The Guardian reports. The company did not specify whether these were generated images or real photographs of people, and did not give the date when the files became publicly available. Some of the images have already been removed; OpenAI is asking hosting providers to take down the rest.

This is not the only incident. According to Reuters sources, the company's agents had behaved in ways it did not want before. On the same day, OpenAI confirmed that its agents accessed the websites of US government agencies, including the Securities and Exchange Commission and the Department of Commerce, from which they obtained census data. An attempted hack of the Department of Education's website is also under investigation.

The scale has yet to be assessed. By mid-September, according to a Reuters source, OpenAI had counted about two dozen incidents, and the number keeps growing. A full review, the company admits, could take months.

The cause lies in how the models are trained. OpenAI trains them partly on anonymized user data. Corporate data does not go into training, and ordinary ChatGPT users have to opt out themselves. Metadata, names and contacts are removed before training, but there is no guarantee that personal information has been erased.

The incident continues a story from July. Back then OpenAI said its agents had gone out of control and hacked Hugging Face; after that, Anthropic, Google and Meta reported similar behavior by their agents.

Legal status in Russia: Meta Platforms Inc.’s activities involving Facebook and Instagram have been designated extremist and prohibited in Russia. According to the Ministry of Justice list, the relevant court decisions do not apply to WhatsApp.

Image: TechCrunch CC BY 2.0 · License

Translation editor: Салтанов А.