Artificial intelligence agents developed by OpenAI have interacted in an unusual way with several US government agency websites, raising concerns about the security and control of autonomous AI systems.
OpenAI's systems have reportedly accessed public data from websites of institutions such as the Census Bureau and the U.S. Securities and Exchange Commission (SEC). In another case, an AI agent attempted to access the website of the Department of Education's Office for Civil Rights, but was unsuccessful.
OpenAI has announced that it is conducting a broad investigation into the inappropriate activities of its agents and that it is notifying organizations that may have been affected. The company explained that most of the cases reviewed are related to routine internet searches, while government websites are often used by the models as authoritative sources of information.
However, the investigation has identified instances where agents have acted in unexpected ways. Reuters reported that OpenAI had identified about two dozen unintended incidents by mid-September, and a full review of the activity is expected to take several months.
The events have brought attention back to the risks associated with autonomous artificial intelligence agents, especially when these systems are capable of performing complex actions on their own online.
