July 8 (Reuters) – OpenAI will publicly launch its most capable model, GPT‑5.6, on Thursday, after delaying the launch last month on the U.S. government’s request amid national security concerns that powerful AI systems could be misused.
This comes on the heels of the U.S. government lifting curbs on Anthropic’s latest Fable and Mythos AI models last week, less than three weeks after the company was ordered to suspend their access over national security risks.
Washington has increased scrutiny of advanced AI model releases to identify potential threats on concerns that the technology could be misused by military or intelligence in China, Russia or other countries of concern.
Axios, which broke the news on the OpenAI launch, reported citing a source familiar with the matter, that the U.S. Department of Commerce had approved a broad launch of GPT-5.6, following additional government testing under Washington’s new oversight framework for frontier AI.
OpenAI had limited the model’s access to a small group of vetted partners whose details were shared with the authorities.
The tech firm now plans to launch GPT-5.6 Sol, along with Terra and Luna models, OpenAI said in a post on X late on Tuesday.
Sol is OpenAI’s most advanced model yet, while Terra is the mid-tier lower-cost model and Luna is the most cost-efficient option.
The White House and the U.S. Department of Commerce did not immediately respond to a Reuters request for comment outside regular business hours.
Increased scrutiny of AI models began with U.S. President Donald Trump signing an executive order establishing a voluntary framework for AI developers to offer “covered frontier models” to the U.S. government for up to 30 days before releasing them to trusted partners.
Anthropic has warned that it was “probably impossible” to make any AI model fully robust to jailbreaks and noted the potential for the development of a universal jailbreak that would be able to unblock “an entire class of harmful behaviors.”
Reporting by Devika Nair in Bengaluru; Editing by Sherry Jacob-Phillips and Mrigank Dhaniwala.
