Uncle Sam Says No: After Anthropic, the US now limits access to OpenAI’s new AI models
OpenAI recently unveiled Sol, its latest GPT-5.6 model, alongside Terra (meant for daily work) and Luna (a more affordable version). Sol is OpenAI’s most powerful model yet and is claimed to be at par with Anthropic’s Mythos. But this announcement came with a development OpenAI was obviously unhappy about. The US government, concerned about the models’ advanced capabilities, had their roll out restricted to a small group of partners it has approved.
OpenAI is the second firm to have its models restricted at the behest of the US government. Earlier, the Trump administration had Anthropic abruptly disable both Mythos (whose access was anyway restricted to a small group) and a version of Mythos with guardrails called Fable 5, when it issued a directive that made it illegal for all foreign nationals, even those working at Anthropic, to access these models. The government has since somewhat eased these restrictions and allowed their access to some ‘trusted’ US organisations.
The US has taken this approach partly to ensure that it retains the edge over other countries in the AI race. But it has also raised concerns that it might land up creating an uncertain regulatory environment that unintentionally benefits countries like China. Interestingly, these developments come at a time when a new Chinese model called Z.ai has been generating headlines and being compared to Mythos in its capabilities. There is still a gap between the frontier models from the US and China, but that might be narrowing.