White House Asks OpenAI to Limit GPT 5.6 Release for Security Reasons

OpenAI logo. (Photo: Yonhap)

JAKARTA - OpenAI's latest model, GPT 5.6, is reportedly not immediately released to the public. The Trump administration is said to have asked OpenAI to restrict access to the model in the early stages for security reasons.

TechCrunch, citing The Information, reported that OpenAI plans to provide access to GPT 5.6 only to a select number of partners during the preview period.

In an internal meeting this week, OpenAI CEO Sam Altman reportedly said the government would approve customer access one by one during the period. If the trial phase goes well, OpenAI hopes to launch a wider public launch a few weeks later.

The new model is also being reviewed by the government. The Information reported that the OpenAI team worked closely with a number of government agencies ahead of its launch. The request for the launch to be carried out in a limited manner is said to have come from the Office of the National Cyber Director and the Office of Science and Technology Policy.

In recent months, the Trump administration has begun pushing for federal oversight of new AI models. Earlier this month, Trump signed an executive order requiring a number of AI companies to voluntarily submit their latest models to the government for testing and evaluation before releasing them to the public.

A similar issue previously emerged at Anthropic. The company garnered attention after announcing its latest cyber model, Claude Mythos, would only be available to a small group of partners through the Project Glasswing program.

Anthropic reasoned that the model has such a powerful ability that it has the potential to be misused if it falls into the wrong hands. A number of observers then debated whether the move was purely driven by security concerns or was also part of a marketing strategy.

According to TechCrunch, quoted Friday, June 26, concerns about advanced AI models are mainly related to capabilities in the field of cybersecurity. Large language models (LLM) or large language models have been shown to be able to create malware, namely malicious software designed to damage or infiltrate computer systems. Some models are even said to be able to carry out ransomware attacks, which are attacks that lock victim data and then ask for a ransom to restore the data.

Cyber models such as Claude Mythos are also feared to be able to find and exploit software security gaps at speeds that human analysts find difficult to match.

Because many software systems still have hidden loopholes that can be a gateway to corporate networks, this ability is considered potentially increasing the risk of cyber attacks on various organizations.

However, because the models have not been opened to the public until now, the actual threat is still difficult to measure.