OpenAI on Friday introduced three new frontier artificial intelligence models, but unlike previous launches, their initial deployment will be limited to a small group of trusted partners as the company complies with a new U.S. government oversight process designed to evaluate the national security implications of increasingly powerful AI systems.
The rollout marks another milestone in the Trump administration’s emerging approach to AI governance, which seeks to balance rapid innovation with concerns that advanced models could dramatically enhance offensive cyber capabilities, biological research, and other high-risk applications.
The ChatGPT developer announced in a blog post that its latest models, GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna, will initially be available only to a select group of partners while government reviews continue. OpenAI said broader public access is expected within the coming weeks.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Although the company did not identify the organizations receiving early access, it stressed that restricting deployment was a temporary measure rather than a long-term policy.
“We don’t believe this kind of government access process should become the long-term default,” OpenAI said. “It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”
The company added that it had shared the capabilities of the models and its deployment plans with the U.S. government before Friday’s launch.
AI Regulation Enters a New Phase
The limited rollout confirms the Trump administration’s increasingly active role in overseeing frontier AI development. Earlier this month, President Donald Trump signed an executive order directing federal agencies to establish a framework for evaluating powerful AI systems before they are broadly released. While the order stopped short of imposing mandatory licensing or approval requirements, it encouraged developers to voluntarily provide government officials with early access to assess the capabilities and potential risks of new models.
OpenAI said it is cooperating with the administration to help establish that process.
“We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks,” the company said.
It added that it is working with the administration to develop a “repeatable process for future model releases,” suggesting the framework could become a standard mechanism for evaluating future frontier AI systems.
The approach represents a notable departure from previous years, when developers typically released new models with minimal government involvement.
The new process follows a high-profile confrontation between Washington and Anthropic earlier this month. Anthropic was forced to suspend international access to its latest Mythos and Fable models after the U.S. government imposed export controls over what officials described as national security concerns surrounding vulnerabilities known as jailbreaks.
The restrictions remain in place while Anthropic continues negotiations with federal officials over security standards and future deployment requirements. That dispute exposed the absence of a common framework for evaluating AI safety and accelerated efforts by both government and industry to establish standardized benchmarks for measuring model risks.
Sol Becomes OpenAI’s Most Capable Model
Among the three newly announced systems, GPT-5.6 Sol represents OpenAI’s flagship offering. According to the company, the model delivers improvements across several advanced domains, including software engineering, scientific reasoning, and biology.
OpenAI also described Sol as its strongest cybersecurity model to date. The company said the model performs significantly better at identifying and fixing software vulnerabilities than at carrying out complete cyberattacks, allowing it to remain below the company’s highest internal risk category.
OpenAI said Sol does not cross its “critical” cybersecurity threshold, defined as introducing “unprecedented new pathways to severe harm.”
The announcement comes as cybersecurity has emerged as one of the principal concerns surrounding frontier AI development, with governments worried that advanced models could enable hackers to discover and exploit vulnerabilities faster than defenders can respond.
Regulators Warn AI Is Accelerating Cyber Threats
Those concerns are extending well beyond the United States. Also on Friday, the president of Switzerland’s financial markets regulator warned that banks and financial watchdogs must rapidly adopt artificial intelligence themselves if they hope to defend against increasingly sophisticated cyberattacks.
Marlene Amstad, president of the Swiss Financial Market Supervisory Authority (FINMA), said AI is fundamentally changing the cybersecurity landscape.
“As hackers move faster, banks must adapt by patching vulnerabilities more rapidly,” Amstad told Reuters.
Her comments followed an international hackathon organized by FINMA and the International Organization of Securities Commissions (IOSCO), which brought together about 100 policy and technology specialists to develop AI-powered supervisory tools for monitoring financial markets, including cryptocurrency trading.
IOSCO represents regulators overseeing approximately 95% of global financial markets, highlighting the scale of international efforts to integrate AI into financial supervision.
Amstad said regulators are also examining whether security safeguards can eventually be embedded directly into digital asset infrastructure rather than relying solely on external monitoring.
The cybersecurity debate is unfolding alongside an increasingly intense international race for AI leadership. Amstad cited recent experience with Anthropic’s Mythos models as evidence that frontier AI can expose new operational vulnerabilities even as it strengthens defensive capabilities.
Despite rising security concerns, Amstad argued that limiting access to advanced models would ultimately weaken cybersecurity rather than strengthen it.
“Switzerland must retain access to the most advanced AI models,” she said, adding that AI will be instrumental in strengthening digital systems before they are deployed.
AI Companies remain under pressure to release more capable systems to maintain technological leadership, while governments are demanding greater visibility into models that could influence cybersecurity, national security, scientific research, and critical infrastructure.
Rather than viewing those objectives as mutually exclusive, OpenAI appears to be taking a position as a partner in developing a new regulatory model, one that allows governments to assess the capabilities of powerful AI systems before public release without permanently restricting access.