OpenAI, the developer of ChatGPT, has announced that it will postpone the full public launch of its latest series of artificial intelligence models, GPT-5.6, in response to a request from the US government. Access in the first phase will be limited to a select number of trusted partners.
In a blog post published alongside the announcement, the company said it had begun a "limited preview" of the GPT-5.6 series, which comprises three models: Sol, Terra, and Luna.
OpenAI said it had reviewed its launch plans and the models' capabilities with the US government prior to making the announcement, adding that, at the government's request, it had begun making the models available to a small group of trusted partners — whose participation the government had been informed of — before a broader rollout.
OpenAI confirmed that it intends to make the models available to all users within the coming weeks, noting that the current restriction is a temporary measure being implemented in parallel with efforts to work with US authorities to establish a permanent security review mechanism for advanced models.
The company stressed that it does not believe granting the government advance access to models should become a permanent procedure, arguing that such a mechanism could deprive users, developers, companies, and cybersecurity experts around the world of access to the latest artificial intelligence technologies.
The decision comes as the US administration tightens its procedures regarding advanced AI models, amid growing concerns over cybersecurity and the potential for misuse of these technologies in cyberattacks or military applications — concerns that have prompted the government to seek a review of certain models before their public release.