OpenAI is delaying the release of its latest artificial intelligence model at the request of the US government, mirroring the launch of its Anthropic Mythos product.
The company behind ChatGPT expressed dissatisfaction with the move, saying it would keep the best AI tools for โthe users, developers, enterprises, cyber defenders and global partners who need them.โ
On Friday, OpenAI said it was launching the GPT 5.6 series with a โsmall groupโ of trusted partners.
โAs part of our ongoing collaboration with the US government, we reviewed our plans and model capabilities ahead of todayโs launch,โ OpenAI wrote in a blog post. โAt their request, we are starting with a limited preview to a small group of trusted partners whose participation has been shared with the government, before releasing a wider version.โ
Anthropic, a close competitor to OpenAI, had a similar release program for its Mythos model, but has now abandoned the technology entirely after the US government ordered the company to block foreign citizens from accessing public versions of the model, which has powerful cyberhacking capabilities. Anthropic initially voluntarily delayed the wide-scale release of Mythos, but OpenAI appears to have done so at the behest of the federal government.
Expressing his concerns about the US governmentโs approach, he said: โWe do not believe that this government access process should become a long-term default.โ
However, the company said it believes the โshort-termโ move is โthe surest path to wider availability in the coming weeksโ as it works with the White House to develop a system to test and implement new models, as required by President Donald Trumpโs executive order.
In another parallel to Anthropicโs conflict with the US government over Mythos, OpenAI said all organizations given access to GPT 5.6 will be based in the US, but the company hopes to add foreign partners next week. However, employees of eligible companies based overseas in โsupported countriesโ, including the UK and Australia, will have access to the model.
GPT 5.6 includes three versions: Sol, the strongest of the pack; Terra, which has slightly less performance capabilities, but is cheaper for users; and Luna, the cheapest version.
OpenAI said GPT 5.6 Sol was its โstrongest modelโ but did not exceed the โcybercritical thresholdโ as part of its internal framework for measuring dangerous AI capabilities. It added that Saul was โbetter at helping people find and fix vulnerabilities than reliably conducting end-to-end attacks.โ
OpenAI Chief Executive Sam Altman told employees this week that the federal government had asked for a phased release. He said the government would โapprove every customerโs access during this preliminary periodโ for GPT 5.6 and that more generally it would be released โa couple of weeks laterโ if the process goes well.
skip the previous promotional newsletter
Free Newsletter | Every weekday
Register in Business today
Get ready for your workday โ every morning we will provide you with all the business news and analysis you need.
after promoting the newsletter
โWe have made clear to the US government that this is not our preferred long-term model and will work with it and others in the industry to achieve a more sustainable approach to future releases,โ Altman said in the note obtained by The Information.
OpenAI worked with the US government on a preliminary version of the model that will be used to power the OpenAI ChatGPT chatbot. The phased release was requested after discussions with two government agencies: the Office of the National Cyber โโDirector and the Office of Science and Technology Policy.
The Information reported that Trumpโs Commerce Secretary Howard Lutnick intervened even against the limited release, calling Altman to seek approval from other agencies.
This month, the US president signed an executive order creating a voluntary framework to allow the federal government to test powerful new artificial intelligence models before they are released.
The executive order represents a departure from the White Houseโs previous stance on AI deregulation. Last year, Vice President J.D. Vance said that โover-regulation of the AI โโsector could kill a transformative industry.โ
However, the White Houseโs position has changed amid rapid improvements in the capabilities of models, including Mythos, which Britainโs artificial intelligence safety body called a โstep forwardโ over previous cutting-edge models.