OpenAI is set to launch its highly anticipated GPT-5.6 series on Thursday after the United States government completed a security review that temporarily delayed the release over cybersecurity concerns.

The rollout, originally scheduled for June, was postponed after the White House requested additional time to assess the potential national security implications of the new AI models. Following technical testing and consultations with OpenAI executives, US officials have now cleared the release.

OpenAI describes GPT-5.6 as its most powerful AI model series to date, featuring significant advancements in cybersecurity, biology, and autonomous (agentic) AI capabilities. The company says the new models represent a major step forward in AI performance while incorporating stronger safety measures to reduce the risk of misuse.

The temporary delay reflected growing concerns in Washington over the increasing capabilities of advanced AI systems. US officials have warned that highly capable models could identify software vulnerabilities that might be exploited by hostile actors, including foreign governments and cybercriminals.

The review followed similar action taken against AI company Anthropic, whose advanced Fable and Mythos models were previously restricted due to national security concerns before those limitations were eased.

Although OpenAI cooperated with the government's request to delay the launch, the company emphasized that such pre-release government reviews should not become a permanent practice. OpenAI stated that routinely withholding advanced AI systems from users, businesses, and researchers could slow innovation and limit access to valuable technologies.

The security review comes amid evolving US AI policy. While President Donald Trump has generally supported a deregulatory approach to artificial intelligence to strengthen America's competitiveness with China, his administration recently introduced a voluntary cybersecurity framework encouraging AI developers to share advanced models with the government up to 30 days before public release.

OpenAI confirmed that it is working with federal officials to help establish a repeatable process for future model releases under the new cybersecurity executive order.

The GPT-5.6 family will launch in three versions:

Sol – OpenAI's flagship and most capable model, designed for advanced reasoning and complex professional applications.
Terra – A balanced mid-range model optimized for everyday productivity and general-purpose tasks.
Luna – A lightweight, cost-efficient model focused on speed, affordability, and high-volume applications.

To address security concerns, OpenAI said GPT-5.6 incorporates a "layered safeguard stack," a comprehensive set of safety mechanisms intended to reduce the risk of misuse in cyberattacks and other harmful activities while maintaining strong performance across a wide range of applications.

The launch of GPT-5.6 is expected to mark another significant milestone in the rapid evolution of artificial intelligence, highlighting both the expanding capabilities of frontier AI models and the growing emphasis on balancing innovation with national security and public safety.