OpenAI to launch GPT-5.6 after government-forced delay

TL;DR:

  • OpenAI will publicly launch GPT-5.6 on Thursday, its most capable model to date.
  • The release was delayed last month at the US government’s request over national security fears.
  • Axios reports the Trump administration approved a broad rollout after further testing.

OpenAI is set to release GPT-5.6 on Thursday, ending a rollout that Washington itself paused. The company had restricted its most capable model to a small group of vetted partners after a government-requested delay in June, amid concern that powerful systems could accelerate sophisticated cyberattacks. According to Axios, which broke the news, the Trump administration signed off on a broad launch after additional testing and meetings with officials.

Frontier releases meet the state

The episode shows how thoroughly national security has entered the model-release process. OpenAI will ship GPT-5.6 Sol alongside lower-cost Terra and Luna variants, having shared details of its vetted partners with authorities. The pattern echoes rival Anthropic, which disabled its Mythos and Fable models globally after a 12 June export-control order and only restored Fable once new safeguards were in place; Mythos, aimed at cybersecurity professionals, remains limited to some “trusted” US organisations. President Trump has since signed an executive order letting developers submit “covered frontier models” to the government for up to 30 days before release, formalising the voluntary standards Washington has been drafting. Elon Musk’s SpaceXAI said it would make its Grok 4.5 model publicly available the same week.

Looking forward

For UK businesses, the practical takeaway is timing and access. Frontier capability is increasingly gated by US security reviews, which means the most powerful tools may reach British users on Washington’s schedule rather than the vendor’s. That backdrop sharpens the case for sovereign AI capability, and it comes as cheaper Chinese models close the performance gap. Anthropic has conceded it is “probably impossible” to make any model fully robust to jailbreaks, a candid admission that the security questions delaying GPT-5.6 will not be resolved by this launch alone.

Share this article