OpenAI is set to launch its GPT-5.6 AI model series on Thursday, weeks after postponing the launch in June following a review of potential security risks by US authorities.
According to Axios, the White House has now approved the launch after technical testing and discussions with OpenAI officials.
OpenAI describes its GPT-5.6 model as the most robust to date, with enhanced capabilities in cybersecurity, biology, and autonomous, or agentic, AI tasks.
Why did the White House delay the release?
The delay comes amid growing concern in Washington over increasingly powerful AI systems and their potential misuse.
In June, the White House ordered OpenAI rival Anthropic to suspend use of its cutting-edge Fable and Mythos AI models over national security concerns. Restrictions on those systems were eased just last week.
US officials have concerns about the ability of advanced AI models to identify vulnerabilities in software code that could be exploited by foreign militaries or intelligence services, including Russia and China.
When the White House requested a delay in OpenAI’s GPT-5.6 launch over similar concerns, the company agreed.
But in a statement, the company said it doesn’t believe “this kind of government access process should become the long-term default” because it “keeps the best tools” away from users, businesses and others who need them.
Washington pushes for AI safeguards
At the beginning of his second term, President Donald Trump largely took a regulatory approach to AI, arguing that excessive regulation could weaken the United States’ ability to compete with China in this area.
But in June, Trump signed an executive order aimed at addressing cybersecurity risks associated with advanced AI models. The measure is supposed to establish a voluntary process whereby AI developers can share models with the government 30 days before public release.
OpenAI said it was coordinating with authorities “to develop a cyber executive order framework and repeatable process for future model releases.”
GPT-5.6 will launch in three versions
According to OpenAI, the GPT-5.6 model series released on Thursday will include three versions. The first, the Soul, is described as the company’s “strongest model yet.” The Terra edition will offer a mid-range model aimed at everyday tasks, while the Luna will serve as the cheapest option focused on speed and efficiency.
The company also said that the GPT-5.6 model includes a “layered security stack” – multiple security mechanisms designed to prevent misuse of AI for cyberattacks and other harmful activities.
Edited by: Natalie Muller