OpenAI released GPT-5.6 Sol on Thursday — but with an unprecedented restriction: only 20 organizations pre-approved by the US government received access. The rollout marks the first time Washington has acted as an explicit gatekeeper for a frontier AI model.

Forbes reports the list includes major defense contractors, select federal agencies, and a handful of academic labs — but no foreign entities and no startups. OpenAI CEO Dario Amodei told CNBC the company "does not agree with these restrictions in principle" but complied to maintain a working relationship with the administration.

TechCrunch adds that the restrictions extend to model weights and fine-tuning access, effectively creating a two-tier AI ecosystem: those inside the circle of trust, and everyone else. The move has drawn sharp criticism from open-source advocates who warn it sets a dangerous precedent for government control over AI access.

Critics argue the rollout reflects a broader shift in how frontier models are governed — moving from corporate self-regulation to direct federal oversight in real time.

Sources: Forbes, CNBC, TechCrunch