According to external media reports, the latest OpenAI model GPT 5.6 will be released in a different way than in the past. Instead of being open to the public for the time being, the company provided a preview to a small number of close partners, as the Trump Government had asked it to slow down the up-to-line pace.
It was reported that Chief Executive Officer Sam Altman of OpenAI had indicated at the internal meeting this week that access to the preview phase would be approved by the Government on a client-by-client basis. If the limited open round goes well, the company hopes to move forward with wider distribution in a few weeks.
Two White House agencies are involved.
It was mentioned that the agencies involved included the Office of the Director General of the United States National Network and the White House Office of Science and Technology Policy. In addition to the review of the model, the OpenAI team has been identified as maintaining close collaboration with the Government in the preparation of the release.
This means that the Trump government's attitude to the front-line AI model is changing. Its early emphasis on less interventional routes towards AI has begun in recent months to promote new model reviews at the federal level.
The new executive order requires a first test.
Earlier this month, Trump signed an executive order requiring selected AI companies to voluntarily submit testing and evaluation to the Government prior to the publication of the new model. This limited GPT 5.6 is seen as an extension of this practice.
From the information currently disclosed, the Government's focus remains on the potential network security risks posed by high-capacity models, in particular their ability to detect loopholes, automate attacks and generate malicious codes.
Frontline network model dispute heating
A similar practice had previously appeared on Anthropic. Earlier this year, the company announced that its front-line network model Claude Mythos was open only to a few partners through a project called Project Glasswing, triggering outside discussions on the boundaries between “security restrictions” and “market packaging”.
It was noted that the Generating AI had proved to be capable of preparing malicious software and that some models could even autonomously implement a complete extortion software attack process. Proponents of restrictions on publication believe that such front-line web tools, once misused, may be identified and used at a much faster rate than manual analysts.
However, as the models are not open to the public, it is still difficult for the outside world to independently determine the real capacity boundaries and whether the associated risks have reached a level that requires a case-by-case review by the Government.