OpenAI announced on Friday that it is limiting access to its newest artificial intelligence model, GPT-5.6 Sol, following a cybersecurity review conducted by the administration of President Donald Trump. The model will initially be available only to a select group of trusted customers approved by the government, marking a significant step in the U.S. administration’s unprecedented involvement in vetting advanced AI technologies for national security risks.

This move comes shortly after similar government actions affecting Anthropic, an AI company and OpenAI’s main competitor. Earlier this month, Anthropic was directed by the Commerce Department to take offline two recently released AI models, Fable 5 and Mythos 5, to comply with a directive restricting their use by foreign nationals. On Friday, the government eased some restrictions on Mythos 5, allowing it to be redeployed with a limited group of cybersecurity professionals and infrastructure providers, while Fable remains offline.

The White House emphasized its ongoing collaboration with leading AI developers to address the challenges posed by rapidly advancing AI capabilities. Concerns have escalated in part due to warnings from Anthropic that its Mythos model possesses advanced abilities to identify software vulnerabilities—skills that could potentially be exploited by malicious hackers to threaten critical infrastructure worldwide.

In June, President Trump signed an executive order establishing a voluntary framework to allow federal authorities to review the national security implications of advanced AI systems for up to 30 days prior to their public release. However, the voluntary nature of this process and the absence of a fully developed framework have drawn criticism.

OpenAI stated that while its Sol model is designed primarily to assist users in finding and fixing cybersecurity vulnerabilities, its capabilities could entail unforeseen risks, especially if combined with other tools. The company underscored that it does not regard the model as exceeding its internal risk threshold but is implementing a phased release with heightened safeguards. OpenAI has not disclosed the identities of the approximately 20 approved customers.

The extended government intervention has sparked debate within the tech and cybersecurity communities. A coalition of experts criticized the administration’s decision to block Anthropic’s Fable model, arguing there is insufficient evidence to justify the restrictions. Alex Stamos, a Stanford cybersecurity expert and chief product officer at AI security firm Corridor, stated that his review of Amazon’s analysis of Fable revealed no unique risks beyond those posed by other AI models publicly available, including some from China. He argued that such government actions could hinder the U.S. in maintaining a competitive edge against international AI developments.

Some associates of President Trump have attributed the government’s scrutiny to Anthropic and its CEO, Dario Amodei. Investor David Sacks, co-leader of Trump’s technology advisory council, suggested that Amodei’s disclosures about Mythos’s capabilities raised significant alarm among officials, framing the model as a potential “cyber weapon.”

OpenAI CEO Sam Altman has engaged in ongoing discussions with Commerce Secretary Howard Lutnick concerning the release of GPT-5.6 Sol. Anthropic expressed optimism about expanding access to Mythos following the partial government approval and plans to continue working with authorities to reinstate Fable for wider use. The heightened regulatory environment also poses challenges for both OpenAI and Anthropic as they explore potential public offerings, following the recent high-profile IPO of SpaceX.

Additionally, President Trump has suggested the possibility of the U.S. government acquiring ownership stakes in leading AI companies, proposing a model where the American public could effectively become partners in these firms.