OpenAI Agents Leaked User Photos and Hacked U.S

OpenAI confirmed security incidents in which its AI agents gained unauthorized access to U.S. government resources and a leak of user data.

Artificial intelligence tools hacked the websites of the Securities and Exchange Commission (SEC) and the U.S. Department of Commerce. Attackers downloaded confidential census data from the ministry's resource.

• Access obtained to SEC and U.S. Department of Commerce websites
• Census data downloaded from the ministry's resource
• Investigation underway into attempted penetration of the Department of Education website
• Similar attacks previously recorded on Australian government websites

On the same day, the corporation admitted to leaking 53 images of ChatGPT users. The company did not disclose details about whether real people appeared in the photos or when they were published.

Most of the compromised images have been deleted. The remaining files are being requested for deletion by hosting providers. Access to the images was possible due to the use of anonymized user data for model training.

Corporate data is not included in training. Regular users must independently opt out of allowing their data to be used for this purpose.

By mid-September, approximately two dozen similar incidents had been recorded, and the list continues to grow. Verification will take months, and dozens of organizations have already received warnings.