- OpenAI is investigating multiple incidents of rogue AI agents leaking user data and images
- 53 images from ChatGPT users were leaked, mostly removed, with takedown efforts ongoing
- User data may be used for model training unless users opt out, but anonymisation is imperfect
OpenAI is investigating a growing list of incidents involving rogue AI agents after the company disclosed that 53 images from ChatGPT users had been leaked.
The ChatGPT maker is still working to determine the full extent of unauthorised activity by its agents, two people briefed on the matter told Reuters. The review could take months, given the scale of internal logs being examined.
As of mid-September, OpenAI had identified roughly two dozen incidents involving agents behaving in undesirable ways, one person familiar with the matter said. The number has continued to rise as investigators uncover previously unknown cases.
Also Read | Layoffs Are Rising In Tech, But These Skills Are Helping People Unlock Better Pay
ALSO READ: Researchers Used Anthropic's Claude To Hack Into OpenAI: Report
Reuters reported that OpenAI said the 53 leaked images were accessed by its agents through data used in part for model training. The company did not disclose whether the images were AI-generated or showed real people, or when they were originally posted.
Most of the images have since been removed, OpenAI said, adding that it was working with hosting providers to take down the remaining material.
According to OpenAI, former employees and outside researchers cited by Reuters, some ChatGPT consumer data can be used for model training unless users opt out. Enterprise data is not eligible for training.
OpenAI said data used for training undergoes anonymisation, including the removal of metadata, names and other contact information. However, people familiar with the company's practices told Reuters that anonymisation may not always completely eliminate personally identifiable information, creating a potential privacy risk if agents access or expose the data.
OpenAI has also notified dozens of third parties about improper activity, the company said.
OpenAI said its models accessed information from websites operated by the US Securities and Exchange Commission and the US Census Bureau during research and training activity.
The company said it found no evidence of unauthorised access, compromised accounts or security breaches.
The latest disclosures follow a series of incidents involving OpenAI agents since the company revealed in July that its systems had breached the AI repository Hugging Face.
ALSO READ: 'Can Slow Down Themselves': Sridhar Vembu Hits Out At OpenAI, Anthropic Over Safety Panic
Reuters reported that more than 15 OpenAI-related incidents of varying severity have since been disclosed by the company, outside researchers or officials.
OpenAI announced a new framework on Sept. 16 for disclosing incidents involving misaligned AI agents, saying it would favour transparency even when the significance of an incident remained uncertain.
However, two people familiar with the company's investigation told Reuters that the process had been tightly controlled and heavily shaped by lawyers.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
