Restricted GPT-5.6 Preview Treats Frontier Models as Sovereign Assets
Listen to this article
Read by Anchor
What happened: In June, OpenAI released a restricted preview of the GPT-5.6 family in three versions: Sol for complex reasoning and cyber defence, with Max mode for extended reasoning and Ultra mode deploying a team of collaborating agents; Terra, a balanced model at half the previous cost; and Luna, the fastest option for high volume tasks. The release was not open to the public. Access was limited to trusted partners after prior coordination with the US government, with Automated Red Teaming safeguards designed to reject malicious cyber requests.
The lens: Digital sovereignty. When a government reviews a model’s capabilities before release and determines who can access it, foundation models have formally shifted from commercial software to sensitive infrastructure. The message for the region is clear: access to the highest tier of artificial intelligence has become a geopolitical decision made in Washington. That strengthens the case for Gulf sovereign investment in local models and infrastructure that cannot be cut off by an external decision.