OpenAI’s GPT-5.6 preview locked behind U.S. government gate; Google reins in Meta’s Gemini appetite

The AI policy story of the weekend is OpenAI’s decision to limit its new GPT-5.6 family to “a small group of trusted partners” after the U.S. government asked it to hold back a broader release ^[1]. The lineup—flagship Sol, balanced Terra, and fast Luna—was announced Friday, but the Trump administration restricted all three, with OpenAI saying preview access is limited to partners “whose participation has been shared with the government” ^[1].

OpenAI made clear it is not happy with the arrangement. “We don’t believe this kind of government access process should become the long-term default,” the company wrote, arguing it keeps advanced tools away from users, developers, cyber defenders, and global partners ^[1]. The move follows the administration’s earlier effective shutdown of Anthropic’s Fable 5 and Claude Mythos 5 after foreign-national access was ordered removed ^[1]. Former White House AI adviser Dean Ball warned that the executive order’s voluntary 30-day review process is creating a de facto involuntary licensing regime that could slow U.S. progress and help China ^[1].

On the model itself, OpenAI says Sol is its strongest yet, with improved agentic coding, biology, and cybersecurity capabilities, plus a “max” reasoning mode and an “ultra” mode that coordinates subagents ^[1]. Sol is priced at $5 per million input tokens and $30 per million output tokens, with Terra and Luna costing half and roughly one-sixth respectively ^[1].

Meanwhile, the AI infrastructure crunch is creating friction between rivals. Google has capped Meta’s use of Gemini models after Meta requested more compute capacity than Google could supply, disrupting some internal Meta AI projects, according to a Financial Times report cited by CNA ^[2]. Google Cloud revenue hit $20 billion in Q1, but CEO Sundar Pichai said compute constraints kept it from growing faster ^[2].

In world news, U.S. and Iranian technical talks in Switzerland concluded with an agreement on working groups and implementation mechanisms, Iran said Tuesday, following the June 17 Islamabad Memorandum of Understanding signed by the two presidents ^[3]. The U.S. agreed to release $12 billion in frozen Iranian assets, which President Trump said would be used to buy American farm goods, and the Treasury issued a 60-day waiver letting Iran sell oil and petrochemicals ^[3]. Tehran also agreed to a communication line with Washington over the Strait of Hormuz, through which about a fifth of global energy exports transit, though Iranian negotiator Mohammad Ghalibaf said the strait “will never return to the way it was before the war” ^[3].

The common thread: even as AI capabilities accelerate, geopolitical gatekeeping—whether Washington’s model-release reviews or Gulf shipping diplomacy—is increasingly shaping who gets access and on what terms.