OpenAI will publicly launch GPT-5.6, its most advanced AI model, on Thursday after a delay last month prompted by U.S. government requests over heightened national security concerns about the potential misuse of powerful AI technologies. The release comes as Washington and Beijing accelerate competing efforts to control access to cutting-edge artificial intelligence that experts warn could dramatically accelerate sophisticated cyberattacks in sectors relying on complex, interconnected and often decades-old technology systems.
OpenAI said GPT-5.6 Sol, Terra and Luna will launch Thursday, following limited access to a small group of vetted partners whose details were shared with authorities. Axios reported that the Trump administration approved a broad launch of GPT-5.6 following additional testing and meetings between the company and government officials. The White House and the U.S. Department of Commerce didn't respond to a Reuters request for comment outside regular business hours.
The Security Calculus
Washington has increased scrutiny of advanced AI model releases to identify potential threats on concerns the technology could be misused by the military or the intelligence establishment in China, Russia and other countries. OpenAI had previewed the models in late June and said GPT-5.6 Sol was competitive with Anthropic's Mythos Preview on the ExploitBench cybersecurity benchmark. The company touted improved agentic capabilities in coding, biology and cybersecurity — precisely the domains that've raised alarm among national security officials.
OpenAI competitor Anthropic had abruptly disabled its most advanced AI models, Mythos 5 and Fable 5, for all users after the U.S. government's June 12 export control order over national security concerns, 26 days ago. The curbs were lifted last week after Anthropic implemented certain safeguards. While Washington has lifted export controls for Anthropic's Fable model, Mythos, which is designed for cybersecurity professionals, is still only available to some trusted U.S. organizations.
A Two-Front Competition
Chinese authorities have also held meetings with top tech firms about potentially restricting overseas access to China's most advanced AI models, including those yet to be released. In China, authorities are worried about the potential for Mythos to exploit software vulnerabilities and that the U.S. might deploy the model against Beijing's interests. The United States and China are in a race to develop cutting-edge AI models that experts have said could dramatically accelerate sophisticated cyberattacks.
Anthropic has warned it was probably impossible to make any AI model fully robust against jailbreaks. Billionaire Elon Musk, whose SpaceXAI rivals Anthropic and OpenAI, said on Wednesday his company was also making its leading model Grok 4.5 available to the public. U.S. President Donald Trump has signed an executive order establishing a voluntary framework under which AI developers could provide covered frontier models to the U.S. government for up to 30 days before releasing them to trusted partners.
Why This Matters:
The delay and approval process for GPT-5.6 marks a significant shift in how governments treat advanced AI as a national security asset rather than purely a commercial technology. The fact that both Washington and Beijing are now holding closed-door meetings with AI companies and implementing export controls suggests the era of unrestricted global AI development is ending. Experts warn that the race to develop ever-more-powerful models capable of exploiting software vulnerabilities creates a dual-use dilemma: the same tools that can identify and patch security flaws can also be weaponized for cyberattacks. The voluntary 30-day government review framework signals that Washington is attempting to balance innovation with security, but Anthropic's own admission that jailbreaks are probably impossible to prevent raises the question of whether any safeguards can truly contain these technologies once they're released. As the U.S.-China competition intensifies, the risk isn't just that one side develops a superior model — it's that the pressure to deploy quickly could outpace the ability to understand what these systems can actually do in adversarial hands.