OpenAI launches new system the Trump administration initially put on a leash
TL;DR
OpenAI broadly launched GPT-5.6 Sol on July 9 after the Trump administration initially pushed the company to limit access to government-approved partners. OpenAI says the US government raised no objections to the public launch, opening the model to companies and individual users. The model is described as especially capable in hacking and cybersecurity tasks, which has pulled Washington deeper into release reviews and testing.
Nauti's Take
The real story is not whether OpenAI had permission, but that everyone is pretending this was just voluntary coordination. Government scrutiny makes sense when a model gets meaningfully stronger at hacking.
But invisible clearance through calls, meetings and political pressure is not a durable regulatory system. OpenAI's safety messaging stays thin until outside testing and clear release criteria are visible to the public.
Briefingshow
This shows how fast AI governance can slide from clear rules into case-by-case power politics. Officially, OpenAI decides its own release timing; practically, the government can slow access when it sees cyber risk. For companies, access to frontier models may depend less on product roadmaps and more on security negotiations in Washington.
Sources