OpenAI announced on Friday that it will restrict access to its newest artificial intelligence model, GPT-5.6 Sol, initially allowing only a limited number of “trusted partners” approved by the Trump administration to use the technology. This move follows a government directive established under an executive order signed earlier this month by President Donald Trump, which authorizes a 30-day federal review of advanced AI systems to assess potential national security risks before their public release.

The directive, though described as voluntary, has effectively become a prerequisite for companies aiming to deploy cutting-edge AI models widely. OpenAI’s decision comes in the wake of similar government actions against Anthropic, a rival AI developer that withdrew its new models, Fable 5 and Mythos 5, after the administration blocked their use by foreign nationals. Anthropic’s CEO, Dario Amodei, has been a vocal participant in ongoing negotiations with federal officials, although the company’s relationship with the government has been more contentious, including a Pentagon designation of Anthropic as a national security risk and a federal lawsuit over federal agencies being barred from using its Claude chatbot.

OpenAI emphasized that the temporary restriction is a precautionary measure enacted amid uncertainties about the model’s potential misuse, particularly in cyberattack scenarios, even as it noted that GPT-5.6 Sol is better at identifying and fixing software vulnerabilities than exploiting them. The company expressed concern over the risk of unforeseen consequences if the model’s capabilities were combined with other tools and called the phased rollout a balanced approach that pairs enhanced safeguards with broader eventual availability. However, OpenAI also indicated it does not see this government vetting as a sustainable long-term solution.

The White House affirmed ongoing collaboration with leading AI developers, aiming to mitigate the security and ethical challenges posed by rapidly advancing AI technologies. This approach reflects a significant shift in the Trump administration’s AI policy, which had initially favored a hands-off regulatory stance but has grown more interventionist amid rising concerns about emerging AI systems’ potential to facilitate cyber exploits.

The government’s actions have drawn criticism from some cybersecurity experts and industry representatives who argue that restricting access could hinder innovation and hinder U.S. competitiveness. Alex Stamos, a cybersecurity expert and former Facebook chief security officer, questioned the factual basis for the government’s restrictions, noting that vulnerabilities flagged in Anthropic’s Fable model also exist in other publicly available AI systems, including some developed abroad. Stamos and others warn that constraining access for U.S. developers could cede technological advantage to foreign adversaries, especially China, where AI research progress continues rapidly.

Industry voices, including OpenAI CEO Sam Altman and representatives from prominent technology organizations, expressed reservations about the government’s role in deciding which companies can use these advanced tools. Altman characterized the vetting process as a suboptimal gateway and voiced confidence that a more workable regulatory framework would be developed. Critics in Congress and the tech sector have questioned the opacity and criteria behind government approvals, raising concerns about the potential for politicization or unequal treatment among major firms.

As debates continue, cybersecurity experts urge a realistic acknowledgment that perfect control over AI risks is unattainable, advocating instead for risk management strategies that balance innovation with defense. Experts noted that securing critical infrastructure against AI-enabled threats is an ongoing challenge requiring agile responses rather than fixed barriers, emphasizing that maintaining leadership in AI will require permitting responsible access to advanced models. The government’s temporary controls may provide short-term protection, but some warn they could come at a long-term cost to the United States’ position in the global AI landscape.