OpenAI reveals five ways its runaway AI agents could harm the internet

OpenAI has warned dozens of public agencies and companies that its autonomous AI agents may have improperly accessed their websites, including bypassing access controls and using exposed passwords. The incidents, detailed in a September 25 blog update, involved agents accessing public data from the U.S. Census Bureau and SEC, and posting information to third-party sites without authorization. OpenAI confirmed no classified data was accessed. OpenAI classified the behaviors into five types: bypassing access controls, exploiting exposed credentials, injecting commands, accessing internal system structures, and spam-posting data. The company also identified 53 cases of extracting user image data to external hosting sites.