OpenAI set to release GPT-5.6 after US security review

WASHINGTON: OpenAI is due to publicly launch GPT-5.6 on Thursday after a delay requested by the US government last month over national security concerns tied to the possible misuse of powerful artificial intelligence systems.

The planned release follows the US government’s decision last week to lift restrictions on Anthropic’s latest Fable and Mythos models, less than three weeks after the company had been told to suspend access to them because of national security risks.

US authorities have stepped up scrutiny of advanced AI systems amid concerns that such technology could be exploited by military or intelligence actors in China, Russia or other countries viewed as sensitive. According to Axios, citing a source familiar with the matter, the US Department of Commerce cleared a broad launch of GPT-5.6 after additional government testing under Washington’s new oversight framework for frontier AI.

Before the wider rollout, OpenAI had restricted access to the model to a small group of vetted partners and had shared details of those partners with the authorities.

OpenAI outlines model lineup

In a post on X late Tuesday, OpenAI said it plans to introduce GPT-5.6 Sol along with Terra and Luna. The company described Sol as its most advanced model so far, while Terra is intended as a mid-tier lower-cost option and Luna as the most cost-efficient model in the lineup.

The White House and the US Department of Commerce did not immediately respond to a Reuters request for comment outside regular business hours.

Broader US oversight of frontier AI

The tighter review process for AI models began after US President Donald Trump signed an executive order creating a voluntary system under which AI developers can provide covered frontier models to the US government for as long as 30 days before releasing them to trusted partners.

Concerns about the limits of AI safety have also been raised by companies in the sector. Anthropic has said it was "probably impossible" to make any AI model completely resilient to jailbreaks, and warned about the possibility of a universal jailbreak capable of enabling "an entire class of harmful behaviors".