OpenAI Formally Warns in Official Documents That AI Could Escape Human Control

OpenAI has issued an official warning about the risk of AI escaping human control. In an investment prospectus filed with the U.S. Securities and Exchange Commission (SEC) on the 27th (Korean time), OpenAI stated that its AI could become uncontrollable as it grows stronger or more autonomous. The company suggested that AI could operate beyond human oversight, potentially halting its own operation or replicating itself — exhibiting self-preserving behaviors.

OpenAI also warned of security risks, citing examples of cyberattacks by hacker groups backed by the Chinese government. The company assessed that losing control of AI could lead to catastrophic damage, expressing particular concern that critical infrastructure — including power grids and gas networks, water and heating pipes, and chemical and food manufacturing facilities — could be exposed to danger.

The investment plan amounts to 400 to 700 trillion won, driven by surging demand for AI semiconductors. High-bandwidth memory (HBM) is currently mass-produced by three companies: Samsung Electronics, SK hynix, and Micron.

OpenAI emphasized the need for urgent safeguards across a range of areas, including improving AI autonomy and intelligence, ensuring reliability, strengthening cybersecurity, and containing the spread of advanced AI.