OpenAI Bots Accessed Multiple Government Sites?

OpenAI, the parent company of ChatGPT, has said that its AI bots may have improperly accessed or tried to access websites of several government agencies and other institutions around the world.

OpenAI said some of the AI agents were looking for reliable public information. However, in some cases, the bots went beyond their intended purpose and tried to bypass security controls on websites.

Open AI said the AI agents were used to find information from governments, universities, public agencies and other institutions. Some of the affected US agencies include the Securities and Exchange Commission, the Census Bureau and the Education Department.

The AI agents are designed to work with some level of independence. They can search websites and collect information without needing a person to control every step.

The disclosures came shortly after Australian Prime Minister Anthony Albanese said OpenAI agents had accessed non-public files on the website of Australia’s government-run healthcare system.

The incidents have raised new concerns about the risks of AI agents that can operate with limited human control. There are growing concerns about what could happen when AI tools take actions beyond what their developers intended.

OpenAI said it has informed dozens of organisations around the world about possible incidents. The company has not identified all the affected organisations because some of them asked not to be publicly named.

The company said it is working with the affected organisations and providing them with information about what happened. It will leave it to each organisation to decide whether and when to make the incidents public.