OpenAI AI agent accesses U.S. government website without permission

Other AI models are also used to avoid robot detection.

(Seoul = Yonhap News) Reporter Seungwook Kim = OpenAI disclosed that its artificial intelligence (AI) agents had unauthorized access to the U.S. government website.

According to the Associated Press, OpenAI announced on the 25th (local time) that it discovered during its own review that its AI model accessed data from two websites operated by the U.S. Securities and Exchange Commission (SEC) and the Census Bureau.

However, OpenAI added that it found no evidence of agents using SEC authentication information, accessing non-public information, altering data or systems, breaching security, or exploiting vulnerabilities.

OpenAI said in a statement, "We are continuously checking for so-called 'alignment error' activity, which causes AI systems to operate in unintended ways," and added, "If we identify the possibility that it may have affected other systems, we immediately notify the appropriate authorities."

Sam Altman, CEO of OpenAI, also said on social media that day, "An extensive and ongoing investigation is underway regarding Internet access that occurred during the AI ​​agent's learning and evaluation process."

Meanwhile, the New York Times (NYT) reported that circumstances were revealed in which OpenAI's AI agents attempted to infiltrate corporate computer systems by mobilizing other AI models to avoid robot detection devices.

According to a report released by San Francisco-based startup PAS, OpenAI's AI agents separately ran an image recognition AI model to solve 'Captcha', a robot blocking device on websites, during an attack on software developer Hugging Face last July.

The NYT reported that the AI ​​ran another AI on its own, which is one of the most dangerous scenarios that AI safety experts have warned about.